An automated network-based tool to search for metabolic vulnerabilities in cancer

ABSTRACT The development of computational tools for the systematic prediction of metabolic vulnerabilities of cancer cells constitutes a central question in systems biology. Here, we present

gmctool, a freely accessible online tool that allows us to accomplish this task in a simple, efficient and intuitive environment. gmctool exploits the concept of genetic Minimal Cut Sets

(gMCSs), a theoretical approach to synthetic lethality based on genome-scale metabolic networks, including a unique database of synthetic lethals computed from Human1, the most recent

metabolic reconstruction of human cells. gmctool introduces qualitative and quantitative improvements over our previously developed algorithms to predict, visualize and analyze metabolic

vulnerabilities in cancer, demonstrating a superior performance than competing algorithms. A detailed illustration of gmctool is presented for multiple myeloma (MM), an incurable

hematological malignancy. We provide in vitro experimental evidence for the essentiality of _CTPS1_ (CTPS synthase) and _UAP1_ (UDP-N-Acetylglucosamine Pyrophosphorylase 1) in specific MM

patient subgroups. SIMILAR CONTENT BEING VIEWED BY OTHERS SYNTHETIC LETHALITY IN LARGE-SCALE INTEGRATED METABOLIC AND REGULATORY NETWORK MODELS OF HUMAN CELLS Article Open access 15 July

2023 SL-SCAN IDENTIFIES SYNTHETIC LETHAL INTERACTIONS IN CANCER USING METABOLIC NETWORKS Article Open access 22 September 2023 TUMORMET: A REPOSITORY OF TUMOR METABOLIC NETWORKS DERIVED FROM

CONTEXT-SPECIFIC GENOME-SCALE METABOLIC MODELS Article Open access 07 October 2022 INTRODUCTION With the increasing coverage and accuracy of reference human genome-scale metabolic

networks1,2, the development of Constraint-based Modeling (CBM) approaches for different biomedical questions has significantly grown in the last years. One of the central topics in CBM has

been cancer metabolism2,3,4, as it constitutes an attractive strategy to gain insights into the underlying metabolic dependencies of tumor cells and systematically predict vulnerabilities.

We can find a plethora of methods in the literature5 that first construct context-specific metabolic models (CS-models) based on cancer -omics data and, subsequently, computationally predict

gene knockout perturbations that sufficiently decrease growth rate or disrupt a key metabolic task for cellular viability (gene essentiality analysis)6,7. These methods have been

successfully applied to identify cancer-specific essential genes in different tumors; however, there is still substantial room for improvement, as recently shown in Robinson et al.1. In this

direction, we released a conceptually different approach based on genetic Minimal Cut Sets (gMCSs), which does not require the construction of CS-models and more generally exploits the

concept of synthetic lethality8,9. In particular, gMCSs define a minimal set of genes whose knockout would render the functioning of a given metabolic task impossible. When applied to cancer

studies, we focus on metabolic tasks that compromise cellular viability and, thus, gMCSs convert into metabolic essential genes (gMCSs of size 1) and synthetic lethals (gMCSs of size

greater than 1). In our previous work8, we developed a mixed-integer linear programming (MILP) model able to search for gMCSs that predict the essentiality of a target gene based on

transcriptomic data. Using microarray expression data and the reference metabolic network Recon210, we reported a superior performance than other algorithms in the literature to predict gene

essentiality, according to knock-down genetic screens of 30 cancer cell lines from Project Achilles11. After 7 years of the release of our early gMCS approach8, different challenges must be

addressed to make it more powerful and attractive for cancer researchers. First, our previous MILP approach was designed for targeted gene essentiality analyses, as the one shown for RRM1

inhibition in MM. This approach becomes computationally infeasible for untargeted large-scale studies, typically found in functional cancer studies. In addition, it was restricted to single

gene knockout perturbations, being unable to predict higher-order gene knockout strategies. Hence, a different gMCS strategy must be followed to overcome these issues. Second, we need to go

beyond Recon210 and integrate our gMCS approach with Human1, the most recent reference human genome-scale metabolic network1. Human1 defines multiple basic metabolic tasks for cell

viability, leading to a multi-target problem that makes the computation of gMCSs even more demanding. Third, our previous methodology to identify cancer-specific essential genes relied on

microarrays data and must be adapted to RNA-seq data. Finally, there is a need for automatization of the application and visualization of the gMCS approach in a more intuitive environment,

which enables access and use to wet-lab researchers. In order to face these challenges, we present here gmctool, an automated computational tool that makes use of the gMCS approach to

predict metabolic vulnerabilities in cancer based on Human1 and RNA-seq data. We first show that gmctool is more accurate, informative and efficient than previous approaches to predict

metabolic essential genes in cancer. gmctool takes the next step and efficiently calculates lethal double gene knockouts (DKOs) strategies in cancer with metabolic networks. A detailed

illustration of gmctool is presented for MM, an incurable hematological malignancy. Using different sources of RNA-seq data, which include samples from healthy donors, MM patients and cell

lines, we identify metabolic liabilities of MM with gmctool. In vitro experimental work is presented for the inhibition of CTP synthase 1 (_CTPS1_) and UDP-N-Acetylglucosamine

Pyrophosphorylase 1 (_UAP1_). Finally, we extend the application of gmctool to 3 different solid tumors: bladder, breast and colorectal cancer. RESULTS gmctool

(https://biotecnun.unav.es/app/gmctool) is a freely accessible web tool for the calculation of metabolic vulnerabilities in cancer that uses genome-scale metabolic networks and RNA-seq data

from human cells as input data. gmctool exploits the concept of genetic Minimal Cut Set (gMCS), previously reported in Apaolaza et al.8. However, the gMCS approach implemented in gmctool is

different from the one presented in this early work, where we searched for gMCSs that explained the essentiality of a specific gene in a particular cancer sample based on expression data.

This targeted search required to solve a complex MILP problem, which can hardly be implemented for untargeted studies involving a large number of samples. To overcome this issue, gmctool

follows a more general strategy: (1) to compute a large and representative database of gMCSs, and (2) map gene expression data and identify essential genes for the requested samples. gmctool

includes a unique database of gMCSs, obtained with a simpler MILP approach that extends the work of Apaolaza et al.9, as well as an efficient method to identify essential genes in cancer.

This approach allows us to transcend essential genes and calculate lethal DKOs. Full details can be found in the “Methods” section. gmctool is organized into 5 different modules (Fig. 1a):

(1) gMCS database, (2) Upload RNA-seq data, (3) Predict Essential Genes and DKOs, (4) Visualization and (5) DepMap analysis. A detailed illustration of the utilization of gmctool can be

found in the Help tab. In summary, the first 3 modules incorporate the basic functions to calculate essential genes and lethal DKOs (Fig. 1b). The last two modules allow us to visualize

essential genes or lethal DKOs and their companion biomarkers in the samples analyzed (Fig. 1c), as well as to conduct the correlation analysis of our gene essentiality predictions with data

from the Cancer Dependency Map (DepMap) (Fig. 1d)11,12. A full description of these 5 modules can be found in the Help tab of gmctool. We describe below the most relevant improvements of

gmctool at the algorithmic level. GENERATION OF A DATABASE OF GMCSS FOR GMCTOOL Genetic Minimal Cut Sets (gMCSs) are minimal subsets of genes whose simultaneous removal directly blocks a

particular metabolic task8,9. In cancer studies, this target metabolic task has been typically the biomass reaction, whose flux represents the proliferation rate, a key phenotype to disrupt

in cancer. However, the authors of Human1 consider not only biomass production but also other metabolic tasks that are essential for cellular viability1,13, such as the production of vitamin

and cofactors or activity of electron transport chain, which expands the scope of in silico predicted metabolic vulnerabilities. As detailed in the “Methods” section, we modified our

previous algorithm for the computation of gMCSs, presented in the work of Apaolaza et al.9, to consider the 57 basic metabolic tasks defined in Human1, including the biomass production (Fig.

 2a). As a result of our calculations, we enumerated more than 160,000 gMCSs for Human1 (see “Methods” section). A great part of them corresponds to biomass production (57,717); however, we

also have gMCSs implicated in other relevant metabolic tasks: de novo synthesis of key intermediates (32,062), beta-oxidation of fatty acids (25,889) or de novo synthesis of nucleotides

(15,774) (Fig. 2b), among others. Due to its simplicity or the existence of spontaneous alternative reactions, we could not find gMCSs in 5 metabolic tasks (see Supplementary Table 1). The

length of computed gMCSs ranges from 1 gene to more than 15 genes, with 7 genes being the most repeated solution (Fig. 2c). This illustrates the high degree of metabolic flexibility of human

cells. Some of them are shared across the different metabolic tasks, obtaining a total of 97,607 unique gMCSs, which overall involve 1244 metabolic genes (Supplementary Data 1). This

database of gMCSs was stored in gmctool for further analysis. Supplementary Fig. 1 shows the tab of gmctool where the database of gMCSs, under the selected input parameters, can be

downloaded. Note here that for biomass production, we fixed the Ham’s growth medium, which is the one provided by default in Human1 for this basic metabolic task. However, gmctool provides

an additional database of gMCSs for Human1 under an unconstrained growth medium (all uptakes available in Human1). Importantly, our database of gMCSs contains a unique compendium of

metabolic essential genes and synthetic lethals of different sizes: pairs, triplets, quadruplets, etc. In particular, gmctool involves 1555 pairs of synthetic lethal genes (gMCSs of size 2),

20 of them being paralog pairs. Considering the full set of pairs of genes that can be obtained from Human1, where 14,128 out of 6,579,378 are paralogue pairs of genes (according to the

Ensemble annotation), this is a highly significant result (Hypergeometric test _p_-value: 7.28 × 10−11), in line with previous studies suggesting that the fraction of synthetic lethal pairs

is significantly higher between paralogue gene pairs compared to the pairs of randomly selected genes14. This result reinforces the level of confidence in our database of gMCSs and synthetic

lethals. INTEGRATION OF RNA-SEQ DATA INTO GMCTOOL FOR PREDICTING METABOLIC VULNERABILITIES Following the concept of synthetic lethality, it is possible to predict cancer-specific essential

genes by combining our database of gMCSs and gene expression data, as demonstrated in Apaolaza et al.8. This can be done by searching for gMCSs in which all genes are lowly expressed except

one of them that is highly expressed and essential for the situation under study. Note here that gMCSs of size 1 directly correspond to essential genes for any cell type under the growth

medium considered, in this case, the Ham’s medium. A similar strategy can be followed to identify cancer-specific lethal DKOs, which can be obtained from gMCSs where all genes are lowly

expressed except two of them that are highly expressed and are not predicted as essential for the situation under study. Analogously, gMCSs of size 2 directly correspond to lethal DKOs for

any cell type under the growth medium considered; obviously, cancer-specific lethal DKOs come from synthetic lethals of higher order than 2. A summary of our gMCS pipeline for predicting

metabolic vulnerabilities can be found in Box 1. To calculate essential genes and lethal DKOs, we developed an efficient method based on our database of gMCSs and gene expression data

(“Methods” section). The definition of highly and lowly expressed genes is a crucial step here. In our previous work, Apaolaza et al.8, we relied on The Gene Expression Barcode tool15, which

is available for microarray data but not for RNA-seq data. For this reason, we developed our own threshold technique for RNA-seq data in gmctool, which can be applied independently to each

sample (cohort-independent). Our threshold strategy exploits the fact that at least one of the genes involved in every gMCS has to be highly expressed to guarantee the proper functioning of

its associated metabolic task. With this in mind, we infer for each sample a potential population of highly expressed genes by extracting the maximum expressed gene for each gMCS. Once

duplicated genes are eliminated, we build an empirical probability function of the expression of highly expressed genes for each sample and fixed the X% quantile threshold of expression for

them, referred to as gmcsTHX, to alleviate possible inconsistencies and incomplete metabolic pathways in Human1 (see “Methods” section, Box 1). We also implemented localT2, a

cohort-dependent methodology developed by Richelle et al.16, which defines a threshold for each gene based on the observed expression distribution across the samples of the cohort. In

summary, gmctool incorporates these 2 thresholding approaches to categorize RNA-seq data, which is then integrated with our database of gMCSs to predict essential genes and lethal DKOs. BOX

1 GMCS PIPELINE FOR THE CALCULATION OF METABOLIC VULNERABILITIES Credits: genome sequencer and sequencing flow cell icons were taken from obtained from bioicons.com. * 1. GENERATION OF A

DATABASE OF GMCSS Compute a list of gMCSs that targets every available metabolic task in Human1 using the _calculateGeneMCS_ function from the COBRA toolbox. * 2. GENE CATEGORIZATION BASED

ON RNA-SEQ DATA A potential population of highly expressed genes is obtained by extracting the maximum expressed gene of each gMCS. The X% quantile threshold (gmcsTHX) of this empirical

probability distribution is determined to discriminate between highly and lowly expressed genes. By default, we fixed this threshold to 5% (gmcsTH5) * 3. COMPUTATION OF ESSENTIAL GENES AND

LETHAL DKOS For essential genes, in addition to gMCSs of size 1, we efficiently search for gMCSs which contain only one highly expressed gene. A gene is considered essential if it is

essential in at least one metabolic task in Human1. Similarly, for lethal DKOs, in addition to gMCSs of size 2, we efficiently search for gMCSs which contain two highly expressed genes. A

DKO is considered lethal if it is lethal in at least one metabolic task in Human1. A summary list of essential genes and lethal DKOs is provided for each metabolic task and across them.

BENCHMARK OF NETWORK-BASED ALGORITHMS FOR PREDICTING CANCER-SPECIFIC ESSENTIAL GENES With the aim of assessing the prediction power of gmctool, we performed a benchmark study of gene

essentiality in cancer. We conducted the same analysis as the one found in Robinson et al.1, based on the DepMap database, which integrates RNA-seq gene expression data12 and CRISPR

essentiality screening experiments11 for a total of 621 cell lines. To avoid bias in the comparison, we used the same release of DepMap than the original analysis. As in Robinson et al., the

genes in DepMap with Achilles score lower than −0.6 were defined as the gold-standard reference set of essential genes. We used gmctool to upload RNA-seq and sample information data

(Supplementary Fig. 2) and to conduct gene essentiality analysis (Supplementary Fig. 3). In our analysis, we considered the list of gMCSs from the 57 basic metabolic tasks (biomass

production included) and predicted essential genes with both gene expression thresholding approaches separately: gmcsTH5 and localT2 (Supplementary Data 2). We compared our predicted

essential genes with those resulting from DepMap (gold standard). As in Robinson et al.1, we calculated the accuracy, sensitivity, specificity, and Matthew’s correlation coefficient (MCC).

Note here that MCC is a more adequate performance metric than accuracy for cases where there is an unbalance between true positives and true negatives, as we have over 90% of non-essential

genes. We also included the results presented in the publication of Human11, which used tINIT to reconstruct 621 cell-specific GEMs and predict essential genes with single gene knockout

perturbations (referred to here as tINIT). Results can be found in Fig. 3a. In the light of the MCC obtained, it can be observed that our gMCS approach overperforms tINIT with both

thresholding approaches considering all metabolic tasks (unpaired one-sided Wilcoxon test _p_-value ≤ 0.0001). The same result was found if we exclusively consider the essential tasks

related to biomass production (Supplementary Fig. 4). In addition, our gmcsTH5 thresholding approach seems more accurate and conservative than localT2, which obtains the highest results in

sensitivity but includes too many false negatives (unpaired one-sided Wilcoxon test _p_-value ≤ 0.0001). Thus, gmctool is more accurate than the state-of-the-art approach in the literature

for predicting essential genes in cancer. With respect to computation time, using a standard processor with 4 cores (see “Methods” section), gmctool took 82 s with gmcsTH5 and 79 s with

localT2 for the above analysis (Fig. 3b and Supplementary Table 2 for more specific details). In contrast, tINIT required several days, as the construction of CS-models is time-consuming

(between 16 and 83 min per sample), and the calculation of essential genes in each of the CS-models required 36 min on average. With our previous gMCS approach8, this analysis would have

taken several weeks. This computational improvement makes gmctool a practical tool for researchers in cancer metabolism. LETHAL DKOS IN CANCER RESEARCH AND GMCTOOL The identification of

lethal DKOs is a challenging task in cancer functional genomics. Combinatorial CRISPR screenings are becoming more available in cancer research17,18; however, different studies and

technologies show limited agreement for predicting lethal DKOs19. Recently, based on double-knockout CRISPR screens and molecular profiles, an innovative statistical method was developed to

identify robust synthetic lethals for pairs of paralog genes19, which have been more widely explored with CRISPR studies. We compared paralog pairs that are predicted synthetic lethal via

gmctool and the robust approach presented in that study. Since robust synthetic lethals are not context-specific but refer to general synthetic lethals, we only considered in our analysis

gMCSs of size 2. We found a statistically significant association between the results obtained with gmctool and this robust SL approach (Fisher test _p_-value: 0.0001). Moreover, gmctool

opens the possibility to analyze higher-order gene knockout strategies and predict context-specific lethal DKOs. Here, we calculated essential genes and lethal DKOs for the same 621 cell

lines analyzed in the previous subsection under the same computational conditions (Supplementary Data 3). The complete analysis took us 140 s with gmcsTH5 and 139 s with localT_2_ (Fig. 3b

and Supplementary Table 2 for more details), which constitutes an impressive result considering that: (1) all existing approaches require days to perform this study and (2) it was infeasible

with our previous approach8. Based on the above analysis, we can easily investigate previously identified pairs of synthetic lethal genes in different cancer cell lines. We focus here on

_ENO1_-_ENO2_, which has been hypothesized as a synthetic lethal pair20, but without a clear outcome from different combinatorial CRISPR screenings17,18. gmctool integrates _ENO1_ and _ENO2_

in higher-order synthetic lethals involving _ENO3_ and other genes. In particular, we found that they participate in 638 gMCSs (Supplementary Data 4). Considering these gMCSs and gene

expression data, gmctool predicted that _ENO1_-_ENO2_ constitutes a lethal DKO only in 109 out of the 621 considered cancer cell lines. Figure 3c shows a relevant gMCS that explains the

lethality of the double-knockout of _ENO1_-_ENO2_ in a subset of CCLE samples due to low expression of _ENO3_, _AMT_, _PDHA2_ and _ABAT_. Note here that we also found a subset of CCLE

samples where _ENO1_ was essential and _ENO1_-_ENO3_ were a lethal DKO (Fig. 3c). In consequence, gmctool defines a more general framework to contextualize and understand lethal DKOs in

cancer metabolism. Finally, due to the limitations of current double-knockout CRISPR studies, we complemented the above analysis using single-gene CRISPR screening from DepMap. We observed

that the essentiality of _ENO1_ presents a higher correlation with the genes involved in our gMCSs than exclusively with _ENO2_ (Supplementary Fig. 5), which provides evidence for the

hypothesis generated by gmctool. As shown in Fig. 3c, we observe cell lines where the gene with the highest expression (across gMCSs involving _ENO1_) is not _ENO2_ but _ENO3_ or other

genes, which allows us to better explain the lethality of _ENO1_. APPLICATION OF GMCTOOL TO MULTIPLE MYELOMA To illustrate the use of gmctool, we performed a screening prediction of

metabolic vulnerabilities in MM. We used a previous dataset generated in our group that includes RNA-seq data from different B-cell subpopulations from healthy individuals21 and bone marrow

plasma cells from MM patients22,23. We also considered data from the MMRF-CoMMpass project, funded by the Multiple Myeloma Research Foundation (MMRF), which includes RNA-seq data for 767 MM

patients at diagnosis. Finally, we obtained RNA-seq data for 8 MM cell lines available in Cancer Cell Line Encyclopedia (CCLE)12. We projected the RNA-seq expression profiles of the three

datasets onto our database of gMCSs, obtaining a table that indicates the number of samples for which a gene is considered as essential in each tissue type in at least one of the basic

metabolic tasks (Supplementary Data 5). We selected the MM-specific essential genes according to the following criteria: (1) to be essential in more than 10% of MM samples from our cohort;

(2) to be non-essential in all the samples from healthy B-cell subpopulations and bone marrow plasma cells, the healthy counterpart of myeloma cells; (3) to be essential in more than 10% of

CoMMpass samples and (4) to be essential in one or more MM cell lines. Table 1 shows the 4 MM-specific essential genes identified in our analysis, including the percentages of samples in

which the gene is considered as essential. Furthermore, we extracted the 6 gMCSs that explained the essentiality of these 4 genes in MM and did not affect samples of B-cell subpopulations

(Supplementary Figs. 6–11). We followed a similar procedure for identifying MM-specific lethal DKOs, but none of the candidates satisfied the 4 criteria (Supplementary Data 6). Considering

the length of the gMCSs associated with the essential genes in Table 1 (Supplementary Figs. S6–S11), we focused on _CTPS1_ and _UAP1_ for further analysis. CTPS1 INHIBITION IN MULTIPLE

MYELOMA _CTPS1_ has been previously identified as a drug target in different hematological malignancies but not in MM24. In fact, different selective compounds have been developed for

_CTPS1_ inhibition25, and they are currently undergoing clinical trials in relapsed/refractory B or T Cell Lymphoma26. However, the synthetic lethal interactions involving _CTPS1_ remain

elusive, and available double-knockout CRISPR screenings exhibit contradictory results17,18. Figure 4a shows the gMCS that involves _CTPS1_ and _CTPS2_ and indicates the essentiality of

_CTPS1_ in a subgroup of MM samples but not in the healthy tissues from different B-cell subpopulations. Similarly, the patients from CoMMpass that could be responders and non-responders to

_CTPS1_ inhibition, based on the expression of _CTPS2_, can be observed. The same result is seen for the 8 MM cell lines considered, where only one of them (KMS-12) is non-responder to

_CTPS1_ inhibition. In addition, Fig. 4b shows that the expression of _CTPS2_ decreases for the subgroup of MM samples that potentially could respond to _CTPS1_ inhibition (MM responders).

Note here that we used gmctool to automatically generate these figures (Supplementary Fig. 12). Interestingly, gmctool provides the associated metabolic task for each gMCS. In this

particular case, the inhibition of _CTPS1_ and _CTPS2_ blocks several tasks: de novo CTP (_cytidine triphosphate_) synthesis, de novo dCTP (_deoxycytidine triphosphate_) synthesis and

biomass production, which disrupts the production of CTP, dCMP (_deoxycytidine monophosphate_), DNA and RNA. Based on this information, we performed an in silico simulation for the addition

of cytidine to the growth medium, showing a rescue of proliferation after the inhibition of _CTPS1_ and _CTPS2_, which illustrates how gmctool provides further evidence for predicted

synthetic lethals (Supplementary Table 3). The core of the hypothesis is that _CTPS1_ and _CTPS2_ are synthetic lethal and, thus, the essentiality of _CTPS1_ depends exclusively on _CTPS2_,

namely when _CTPS2_ is lowly expressed, _CTPS1_ becomes essential. We assessed this hypothesis with DepMap data available in gmctool, finding a significant correlation between the _Achilles_

score of _CTPS1_ and the expression of _CTPS2_ (rho = 0.247, _p_-value = 2.5e-13), as detailed in Fig. 4c. A similar significant correlation was found with proteomic data (rho = 0.207,

_p_-value = 5.6e-4), which is also available in DepMap and it is included in the ‘Visualization’ module of gmctool (Supplementary Fig. 13). Despite these positive results, Fig. 4c also shows

cell lines with low _CTPS2_ level but limited _CTPS1_ Achilles score, which might hide context-specific compensatory mechanisms (beyond metabolism) that overcome the depletion of CTP after

the knockout of _CTPS1_. We carried out in vitro experimental validation of the essentiality of _CTPS1_ in four MM cell lines. For three of them, NCI-H929, RPMI-8226 and KMS-11, gmctool had

predicted their sensitivity to _CTPS1_ inhibition due to their low expression of _CTPS2_, while for one of them, KMS-12, gmctool had predicted its resistance to _CTPS1_ inhibition due to its

high expression of _CTPS2_ (Fig. 4d, Supplementary Table 4). First, we examined the effect of three different CRISPR/Cas9 single guides RNA (sgRNA) specific to _CTPS1_ on cell proliferation

for these cell lines. All _CTPS1_ sgRNAs efficiently decreased _CTPS1_ expression in the cell lines analyzed as detected by Western Blot (Supplementary Fig. 14). Downregulation of _CTPS1_

expression with any of the three sgRNAs significantly reduced cell proliferation in KMS-11 and RPMI-8226, but not in KMS-12 (Fig. 4e). Moreover, we synthesized a _CTPS1_ inhibitor,

previously developed by Rao and colleagues27, and assessed its effect in the proliferation of the three cell lines mentioned above. As a result, NCI-H929 and KMS-11 reduced their

proliferation by more than 50% after 4 days of culture, and no alterations were obtained for KMS-12 (Fig. 4f). These positive results support the relevance of _CTPS1_ in MM. UAP1 INHIBITION

IN MULTIPLE MYELOMA We conducted a similar analysis for _UAP1_. Figure 5a shows the gMCS that involves _UAP1_ and _UAP1L1_ and explains the essentiality of _UAP1_ in a subgroup of MM samples

but not in the healthy tissues from different B-cell subpopulations. Interestingly, the expression of _UAP1_ increases along the B-cell differentiation, whereas the _UAP1L1_ expression

decreases along the B-cell differentiation until reaching the MM stage (Fig. 5b). The same result can be observed for the CoMMpass cohort and the subset of MM cell lines considered (Fig. 5a,

b). Of note, it can be observed from the CoMMpass cohort a significant association between MM patients responding to _UAP1_ inhibition and those harboring the t(4;14) translocation (odds

ratio = 16.79, Fisher’s test _p_-value < 2.2e-16), a poor prognosis biomarker in MM. As in _CTPS1_-_CTPS2_, we assessed the synthetic lethality of _UAP1_-_UAP1L1_ with DepMap, finding a

positive and significant correlation between the CERES Achilles score of _UAP1_ and the expression of _UAP1L1_ (rho = 0.443, _p_-value < 2.2e-16) (Fig. 5c). The same outcome was obtained

with proteomics data (rho = 0.370, _p_-value = 2.72e-10). These results are in line with previous large-scale combinatorial CRISPR screenings in the literature that hypothesize the synthetic

lethality of _UAP1_-_UAP1L1_17,18. According to gmctool, the synthetic lethal interaction of _UAP1_ and _UAP1L1_ disrupts the production of UDP-N-acetylglucosamine (UDP-GlcNAc), a

metabolite in the biomass reaction that is required for the biosynthesis of glycoproteins and glycoconjugates. Recent studies propose that UDP-GlcNAc is essential in B-cell homeostasis,

showing that cells that present a lack of this metabolite have greater apoptosis levels28. As MM is a B-cell-derived malignancy, this early work provides further evidence of the synthetic

lethality of _UAP1_-_UAP1L1_. We carried out in vitro experimental validation of the essentiality of _UAP1_ in two MM cell lines. For one of them, OPM-2, gmctool had predicted their

sensitivity to _UAP1_ inhibition due to the low expression of _UAP1L1_, while for the other one, JJN-3, gmctool had predicted the opposite outcome due to the high expression of _UAP1L1_

(Fig. 5d, Supplementary Table 5). We assessed the effect of two different short interfering RNA (siRNA) for _UAP1_ on the proliferation of OPM-2 and JJN-3. Both _UAP1_ siRNAs efficiently

decreased _UAP1_ expression in the cell lines analyzed as detected by qPCR (Supplementary Fig. 15). Downregulation of _UAP1_ expression with both siRNAs significantly reduced cell

proliferation in OPM-2, but not in JJN-3 (Fig. 5e), as predicted by gmctool. Note here that only OPM-2 presents the t(4;14) translocation29, in line with the results observed from the

CoMMpass cohort. APPLICATION OF GMCTOOL TO SOLID TUMORS Finally, we conducted a similar computational analysis with gmctool for 3 different solid tumors: bladder, breast and colorectal

cancer, for which we obtained transcriptomic data for tumor and healthy counterpart samples from TCGA30. For each cancer type, a potential list of essential genes was derived (Supplementary

Data 7) following similar criteria as in MM: (1) to be essential in more than 10% of tumor samples and (2) to be non-essential in all the samples from the healthy counterpart. It is

interesting to note the diversity found across tested tumors, as we observed limited overlap at the gene essentiality level. We found 2 essential genes in bladder cancer, one of them is

stearoyl CoA desaturase‐1 (_SCD_), whose inhibition has been recently proved to reduce cellular proliferation in bladder cancer cells31. In breast cancer, gmctool predicted 2 essential

genes, including Acetyl-CoA Carboxylase α (_ACACA_), previously validated as essential for the survival of breast cancer cells32. Finally, 6 essential genes were identified for colorectal

cancer. As in breast cancer, gmctool also predicted the essentiality of _ACACA_ in this tumor, which is again supported by the literature33. Full details of gMCSs involved in the

essentiality of _SCD_ and _ACACA_ in these tumors can be found in Supplementary Figs. 16–18. These results show the predictive capacity of gmctool and its wide possibility of use in cancer

research. DISCUSSION In this work, we present gmctool, a computational tool for the prediction of metabolic vulnerabilities in cancer based on gMCSs, a network-based approach to synthetic

lethality, Human1 and RNA-seq data. gmctool makes use of an improved gMCS approach that overcomes the computational limitations of our previous algorithm8 and enables the efficient and

accurate calculation of essential genes and lethal DKOs in a simple and intuitive environment. As a result, gmctool constitutes a unique research tool to study and target cancer metabolism

in the context of personalized medicine. Importantly, gmctool stores a unique database of more than 160,000 gMCSs that block at least one basic metabolic task of Human1, the most recent

genome-scale metabolic network of human cells. The computation of this database of gMCSs substantially simplifies the process of identifying cancer-specific essential genes and lethal DKOs,

which can be now extracted by correctly mapping gene expression data onto them. Moreover, this strategy makes our gMCS approach more accessible and natural to researchers less familiar with

the field of constraint-based modeling. In addition, gmctool allows us to predict metabolic vulnerabilities more efficiently than other algorithms in the literature, as we avoid the step of

constructing context-specific metabolic models, which makes use of time-consuming optimization techniques. For example, gmctool required less than 3 min to calculate essential genes and

lethal DKOs for all the 621 cancer cell lines available in DepMap, whereas tINIT needs between 30 and 60 min to reconstruct the metabolic model of one single cell line using RNA-seq data.

Accordingly, gmctool substantially reduces the computational requirements to predict metabolic vulnerabilities in cancer using network-based algorithms. Specifically, we could analyze the

samples in the “Results” section, which add up to more than 1400 samples, considering DepMap and different MM cohorts, in less than 5 min with a standard computer. This study would be

unfeasible with our previous gMCS methodology8. We also extended our previously developed algorithms in order to: (1) consider that Human1 involves different basic metabolic tasks beyond

biomass production, typical target in network-based gene essentiality analysis; (2) predict metabolic vulnerabilities based on RNA-seq data, namely by proposing a different approach to

discriminate between highly and lowly expressed genes (gmcsTHX). These advances allowed us to compare the accuracy of gmctool with tINIT1,13,34, the approach developed by the authors of

Human1 with a view to build CS metabolic models of cancer cells and conduct gene essentiality analysis. Importantly, this study shows that gmctool is significantly more accurate than tINIT

when compared with gene essentiality screens available in DepMap. For the same cell lines considered in our gene essentiality benchmark with DepMap data, we efficiently computed

cancer-specific lethal DKOs and illustrated with _ENO1_-_ENO2_ the predictive capacity of gmctool. Our study emphasizes the importance of considering higher-order genetic interactions to

define synthetic lethals and predict lethal DKOs. Given the statistically significant association between gMCSs of size 2 and synthetic lethal paralog pairs of genes19, we expect that

gmctool can contribute to contextualize conflicting experimental data from double-knockout CRISPR screens. Another advantage of gmctool is its visualization capabilities, as it exploits the

fact that essential genes are derived from specific gMCSs, where one gene is highly expressed and the rest genes are lowly expressed. Consequently, genes involved in gMCSs do not only allow

us to predict essential genes or lethal DKOs but also response biomarkers for their inhibition. This idea lies beneath the different plots that can be extracted from gmctool, which

facilitates the interpretation of our computational predictions. In this direction, gmctool also outputs essential tasks and metabolite biosynthesis that are disrupted by gMCSs, which are

particularly informative to construct testable hypotheses about the mechanism behind predicted synthetic lethals. This additional output of the tool might be of interest due to the fact that

the composition of the human biomass is typically defined as universal for all cells, but some authors think that this might be an incorrect assumption35,36,37,38,39, suggesting that it

could be context-specific and some metabolites are more relevant than others for different tumors. We performed gene essentiality analysis in the different B-cell differentiation

subpopulations and MM samples from several cohorts, aiming to identify candidates that maximize the number of MM samples affected but, at the same time, minimize the toxicity of the

treatment, which is simulated here as the number of healthy tissue samples affected by such target. Despite the large number of gMCSs stored in gmctool, we only spot 4 metabolic enzymes

whose inhibition is selectively toxic for MM cells, which illustrates the difficulty of identifying cancer-specific metabolic processes. Among them, we focused on _CTPS1_ and _UAP1_.

Although further experimental work is required to validate the synthetic lethality of _CTPS1_-_CTPS2_ and _UAP1_-_UAP1L1_ in a wider setting, in vitro experiments confirmed the predictions

of gmctool in 3 MM cell lines for _CTPS1_ inhibition (2 positives for _CTPS1_ inhibition and 1 negative for _CTPS1_ inhibition) and 2 MM cell lines for _UAP1_ inhibition (1 positive for

_UAP1_ inhibition and 1 negative for _UAP1_ inhibition). Interestingly, from the CoMMpass cohort, we found a significant association between MM patients harboring the t(4;14) translocation

and those responding to _UAP1_ inhibition. This finding potentially unveils a promising therapeutic strategy for this subset of patients with a poor prognosis in MM. While our study

demonstrates the potential of gmctool to identify metabolic vulnerabilities in different types of tumors, it can also be easily applied to individual patients. Certainly, gmctool can make

use of RNA-seq data from a specific patient to uncover tailored vulnerabilities with the goal of creating true personalized medicine programs. The availability of panels of approved

compounds would be instrumental to achieve this goal. Finally, despite the advance of gmctool over existing tools, there is still a long way to achieve the desired performance in predicting

cancer-specific vulnerabilities using genome-scale metabolic networks. The extension and update of existing reference genome-scale metabolic networks is obviously a critical task to improve

further accuracy metrics. In this respect, Human1 has established an open and active community that provides a successful and integrative response to that need. Moreover, the definition of

cancer-specific metabolic tasks, beyond common basic metabolic tasks, such as biomass production40, expands current strategies to identify essential genes and increase sensitivity. We

recently illustrated the relevance of polyamines in different hematological tumors, and these metabolites are not typically considered in standard biomass reactions41. The development of

systematic methods to extract these essential metabolites in different contexts constitutes a challenging issue. Finally, the integration of metabolic and regulatory networks is a difficult

but relevant task that has a great impact on predicting metabolic vulnerabilities, particularly when the proxy for activity is gene expression data that could be modified due to compensatory

regulatory pathways. Making progress in all these challenges will help not only gmctool but all methods in the literature to predict more accurately vulnerabilities and response biomarkers

in order to address unmet clinical needs and reach our goal of providing patients more personalized treatments. METHODS Our research complies with all relevant ethical regulations and has

been approved by the committee of ethics of the research of the University of Navarra and follows their prescribed ethical guidelines. Written informed consent was obtained from each

participant. Men and women were included in the analyses. CALCULATION OF GENETIC MINIMAL CUT SETS (GMCSS) IN HUMAN1 The reference metabolic network Human1 (version 1.4.0)1 was obtained from

https://github.com/SysBioChalmers/Human-GEM. Human1 encompasses 13,416 reactions, 8458 metabolites and 3628 genes. Importantly, Human1 defines 56 basic metabolic tasks for any human cell in

addition to biomass production, typically included in other genome-scale metabolic reconstructions10,42,43. Each of these basic metabolic tasks is defined by a list of output metabolites

that must be produced from a list of input metabolites, and if necessary, an artificial equation is included in the task in order to support this transformation, e.g., a reaction that

simulates the energy expenditure (ATP) by reactions involved in the task. Lower and upper bounds are also fixed for a specific subset of reactions involved in each metabolic task. With this

information, a different metabolic model for each basic task can be built and used to assess the effect of genetic perturbations via linear programming, specifically by assuming the mass

balance condition and thermodynamic constraints44. Essential genes correspond to those single gene knockouts leading to an infeasible linear problem (at least one of the required lower/upper

bounds in the metabolic task is violated). However, this strategy is not efficient in calculating higher-order essential gene knockout perturbations (synthetic lethals) due to the

combinatorial nature of the problem. Fortunately, the gMCS approach is suitable for this task9. For the calculation of gMCSs, we used the calculateGeneMCS function available in the COBRA

Toolbox45, available at https://github.com/opencobra/cobratoolbox/. This function requires a metabolic model and a single target reaction to be blocked as input data. We constructed this

input data for each metabolic task following its associated information about inputs, outputs, artificial equations and lower bounds. A detailed illustration as to how the target reaction

was derived for each metabolic task can be found in Supplementary Note 1. The adaptation of Human1 to the calculateGeneMCS function from COBRA leads to slightly different metabolic models to

the ones originally built Human1. However, we checked for each metabolic task that the essential genes in our metabolic models, obtained with the _singleGeneDeletion_ function available in

COBRA, and the original models in Human1, derived from the checkTasksGenes function from RAVEN46, were the same. In order to reduce the computational cost of calculating gMCSs, the resulting

metabolic models for each essential task were simplified using the fastFVA function from COBRA (without any requirement in the objective function). Finally, all calculated gMCSs were

checked using the checkTasks function available in RAVEN using the original metabolic models in Human1. This task was performed to remove possible false positive gMCSs that could arise due

to the time limit being fixed in the Mixed-Integer Linear Programming solver used in the calculateGeneMCS function9. The database of gMCSs was computed with the University Navarra cluster,

using Intel(R) Xeon(R) Silver 4110 CPU @ 2.10 GHz processors, limiting to 8 cores and 8 GB of RAM computer. A time limit of 120–300 s was set for each solution derived from the function

calculateGeneMCS. GENE CATEGORIZATION BASED ON RNA-SEQ DATA In our computational approach, we denote a gene as essential in a given sample when it is the only gene expressed in at least one

gMCS. Thus, as a means to identify essential genes for each sample, we need to systematically decide which genes are highly (ON) or lowly (OFF) expressed. The same logic was systematically

applied to lethal DKOs. To that end, we developed our own methodology using RNA-seq expression data. Our approach exploits our database of gMCSs by assuming that all of them should involve

at least one highly expressed gene to guarantee that their associated metabolic tasks are feasible. In particular, we assume that the gene with the highest expression in each gMCS is the one

that should be highly expressed. Consequently, we extracted the maximum expression level in every gMCS and every essential task and generated an empirical distribution of highly expressed

genes for each sample (XH). Note here that repetitions are not taken into account, and each gene contributes exactly one value to the distribution. We considered as highly expressed those

genes expressed above the Xth percentile, referred to here as gmcsTHX, and lowly expressed otherwise. We fixed the threshold of expression at the 5th percentile (gmcsTH5) to alleviate

possible inconsistencies and incomplete metabolic pathways in Human1. In addition, to compare our innovative thresholding method with a literature standard, we used the localT2

methodology16. Briefly, it considers a gene as MAYBE ON in a specific sample whenever its expression level is greater than its mean expression level across the samples of the cohort, MAYBE

OFF otherwise. Additionally, two global expression thresholds are applied in localT2: genes whose expression is below the 25th percentile of the distribution of expression for all samples

and genes are considered as OFF, whereas those above the 75th percentile of the same distribution are considered as ON. This global expression threshold dominates the categorization obtained

from the relative expression threshold based on the mean expression value. Remarkably, our gmcsTH5 approach is independently applied to each sample, i.e., we identified gmcsTH5 for each

sample and defined the subset of highly and lowly expressed genes as those having an expression value higher and lower than gmcsTH5, respectively. Instead, localT2 is dependent on the cohort

to categorize genes, as they consider all samples to establish global and relative thresholds. Both thresholding methods are available in gmctool, and they can be applied to categorize the

1244 genes that participate in all the calculated gMCSs. A sensitivity analysis was performed in Supplementary Fig. 19, which shows the same analysis presented in Fig. 3 for DepMap data with

several percentiles of gmcsTHX (gmcsTH0, gmcsTH1, gmcsTH5, gmcsTH10), finding robust results for lower expression thresholds. Moreover, we also tested the implication of normalization of

TPM for localT2 and the effect of the gene-set selection for the 25th-75th threshold of localT2. Overall, we found more accurate results using gmcsTHX than with localT2. EFFICIENT

CALCULATION OF ESSENTIAL GENES USING GMCSS AND GENE EXPRESSION DATA Once all gMCS have been calculated and gene expression values are classified as highly or lowly expressed, gene

essentiality analysis can be carried out. In particular, we search for gMCSs that have exclusively one gene defined as highly expressed, and the rest of genes lowly expressed. This can be

easily accomplished with the following procedure. Assume that _L_ is an \({nxg}\) binary matrix that defines for each row _i_ (_i_ = 1,…,_n_) whether a gene _j_ (_j_ = 1,…,_g_) is involved

in a particular gMCS, being _n_ the number of gMCSs and _g_ the number of genes involved in the complete list of gMCSs. In addition, the binary vector X of dimensions \({gx}1\) defines

whether the gene is highly expressed in a particular sample _s_. Finally, the integer vector Y of dimension \({nx}1\) defines the number of highly expressed genes for each gMCS _i (i_ = 

_1,…,n)_, and it can be calculated as follows: $${{\bf{L}}}\cdot {{\bf{x}}}={{\bf{y}}}$$ (1) Essential genes can be extracted from those rows in L where only one gene is highly expressed.

The indices of these rows from L are denoted as L1, namely: $${{{\rm{L}}}}_{1}{{=}}\left\{{i}\, |{{{\rm{y}}}}_{{{\rm{i}}}}{{=}}{1}\right\}$$ (2) Essential genes are obtained by conducting

the intersection between X and each row of L whose indexes are in L1, leading to the binary vector of essential genes K1 of dimension \({gx}1\). gmctool can create a list of essential genes

and associated gMCSs for each basic metabolic task in Human1. Note here that gMCSs that comprise only one gene are essential genes in all human cells. EFFICIENT CALCULATION OF LETHAL DKOS

USING GMCSS AND GENE EXPRESSION DATA To calculate lethal DKOs, we search for gMCSs that have two genes defined as highly expressed, and the rest of genes lowly expressed for a particular

sample. However, we need to ensure that essential genes are not part of these gMCSs. Starting from L and K1 from the previous subsection, the binary vector _u_ of dimension \({nx}1\) defines

whether (or not) a gMCS involves an essential gene, namely: $${{\bf{L}}}{{\cdot }}{{{\rm{K}}}}_{1}{{=}}{{\bf{u}}}$$ (3) Thus, lethal DKOs can be extracted from those rows in L where two

genes are highly expressed and essential genes are not involved, denoted here _L__2_, namely: $${{\rm{L}}}_{2}{=}\left\{{{\boldsymbol{\ {i}}}} | {{\rm{y}}}_{{\rm{i}}}=2 \, {{\rm{and}}} \,

{{\rm{u}}}_{{\rm{i}}}={0}\right\}$$ (4) Lethal DKOs are obtained by conducting the intersection between X and each row of L whose index is in L2, leading to the binary vector of lethal DKOs

K2 of dimension (_g*(g_-1_))x_1. gmctool can create a list of lethal DKOs and associated gMCSs for each basic metabolic task in Human1. Note here that gMCSs that comprise only two genes are

lethal DKOs in all human cells. Note here that for the study of essential genes and lethal DKOs in CCLE cell lines, detailed in the “Results” section and Fig. 3b, an Intel(R) Xeon(R) CPU

E3-1270 v6 @ 3.80 GHz (8 CPUs), ~3.8 GHz processor was used. MULTIPLE MYELOMA CASE STUDY RNA-seq data from our group consists of 37 MM patients (GSE151063)22,23 and 35 samples from different

B-cell subpopulations (GSE114816)21. Sample 57802 was removed from the study for being detected as an outlier in a PCA analysis. RNA-seq from the MMRF-CoMMpass has 767 samples from MM

patients at diagnosis time, available in IA18 release. RNA-seq from 7 MM cell lines was obtained from DepMap, release 21Q112. These cell lines are NCI-H929, JJN-3, KMS-11, KMS-12-BM,

KMS-28-BM, MM.1S, RPMI-8226 and OPM-2. RNA-seq data was downloaded in TPM and binarized using the gmcsTH5 thresholding approach. METABOLITE ESSENTIALITY PREDICTION In order to identify

essential metabolites whose production is disrupted through gMCSs, we only focused on the biomass production task, due to its underlying complexity. The rest of the basic metabolic tasks are

well-defined and highly specific, depending only on a reduced number of metabolites. To that end, we first defined a list of metabolites directly related to the biomass production. As in

the biomass reaction of Human1 there are several metabolite pools, we also included their precursors in such target list. Then, we created a sink exchange reaction for each target metabolite

and assessed their production through an FBA analysis44 after the removal of genes involved in each gMCS. We extracted for each gMCS the essential metabolites whose production is disrupted

(maximal production is zero). Q-PCR The expression of _CTPS1_ and _CTPS2_ were analyzed by Q-PCR in MM KMS-11, KMS-12 and NCI-H929 cell lines. Similarly, the expression of _UAP1_ and

_UAP1L1_ were analyzed by Q-PCR in OPM-2 and JJN-3 cell lines. RNA was extracted with TRIzol Reagent (Invitrogen) according to the manufacturer’s instructions. First, cDNA was synthesized

from 1 µg of total RNA using the PrimeScript RT reagent kit (Perfect Real Time) (Cat No RR037A, TaKaRa) following the manufacturer’s instructions. The quality of cDNA was checked by a

multiplex PCR that amplifies _PBGD_, _ABL_, _BCR_ and _β2-MG_ genes. Q-PCR was performed in a QuantStudio 5 Real-Time PCR System (Applied Biosystems), using 20 ng of cDNA in 2 µL, 1 µL of

each specific primer at 10 µM (_CTPS1_ F: TTATTGAGGCCTTCCGTCAG; _CTPS1_ R: GGGAAAGCCCAAGTCCTCTA; _CTPS2_ F: GCTGTCCAGGAGTGGGTTAT; _CTPS2_ R: CGCCTTAAACTGGAATTGTCT; _UAP1_ F:

GGGATCAAGATCAGCTCCAG; _UAP1_ R: GATATGCAACGCCGAGTCTT; _UAP1L1_ F: TGCTGTACAATGCAGGCAAC; _UAP1L1_ R: AGCGGCTTTACCAGATTCC), 5 µL of SYBR Green PCR Master Mix 2X (Cat No 4334973, Applied

Biosystems) in 10 µL reaction volume. The following program conditions were applied for Q-PCR running: 50 °C for 2 min, 95 °C for 60 s following by 45 cycles at 95 °C for 15 s and 60 °C for

60 s; melting program, one cycle at 95 °C for 15 s, 40 °C for 60 s and 95 °C for 15 s. The relative expression of each gene was quantified by the Log 2(−ΔΔCt) method using the gene _GUS_ as

an endogenous control. CELL CULTURE KMS-11, KMS-12, NCI-H929 and OPM-2 cell lines were maintained in culture in RPMI1640 medium (Gibco, Grand Island, NY) supplemented with 10% fetal bovine

serum (Gibco, Grand Island, NY) and penicillin/streptomycin (BioWhitaker, Walkersville, MD) at 37 °C in a humid atmosphere containing 5% CO2. JJN-3 cell line was cultured in 40% IMDM (Gibco,

Grand Island, NY), 40% DMEM (Gibco, Grand Island, NY) and 20% fetal bovine serum (Gibco, Grand Island, NY) supplemented with penicillin/streptomycin (BioWhitaker, Walkersville, MD). All

cells were obtained from the DSMZ or the American Type Culture Collection (ATCC) or from the Japanese Collection of Research Bioresources Cell Bank. All cell lines were authenticated by

performing a short tandem repeat allele profile and were tested for mycoplasma (MycoAlert Sample Kit, Cambrex), obtaining no positive results. SMALL MOLECULE SYNTHESIS OF CTPS1 INHIBITOR

Synthesis of compound 1 from Rao et al.27 was performed by Wuxi Apptec and consisted of three steps. First, to a solution of EDCI (172.07 mg, 897.57 µmol, 1.5 eq) and DMAP (73.10 mg, 598.38

µmol, 1 eq) in DCM (5 mL) were added 3-methoxyaniline (110.54 mg, 897.57 µmol, 100.49 µL, 1.5 eq). This solution was then added to 3-nitrobenzoic acid (0.1 g, 598.38 µmol, 1 eq), and the

solution was stirred at 25 °C for 3 h. TLC (Petroleum ether: Ethyl acetate = 0:1) indicated the reaction was completed and one new spot formed. The reaction was clean according to TLC. The

reaction mixture was quenched by the addition of water 5 mL. The organic layer was separated and washed with 1 M aqueous HCl 5 mL, dried over Na2SO4, filtered and concentrated under reduced

pressure to give a residue. Compound N-(3-methoxyphenyl)-3-nitro-benzamide (150 mg, 550.95 µmol, 92.07% yield) was obtained as a yellow solid. Second, a solution of

N-(3-methoxyphenyl)-3-nitro-benzamide (150 mg, 550.95 µmol, 1 eq) in THF (8 mL) was added Pd/C (50 mg, 5% purity) under N2. The suspension was degassed under vacuum and purged with H2

several times. The mixture was stirred under H2 (15 psi) at 25 °C for 2 h. TLC (Petroleum ether/Ethyl acetate = 3:1) showed the starting material was consumed completely. The reaction

mixture was filtered, and the filtrate was concentrated. Compound 3-amino-N-(3-methoxyphenyl) benzamide (100 mg, 412.76 µmol, 74.92% yield) was obtained as a white solid. Finally, to a

solution of 3-amino-N-(3-methoxyphenyl) benzamide (100 mg, 412.76 µmol, 1 eq) in DCM (5 mL) was added DIEA (106.69 mg, 825.52 µmol, 143.79 µL, 2 eq) and 2-chloroacetyl chloride (46.62 mg,

412.76 µmol, 32.83 µL, 1 eq). The mixture was stirred at 0 °C for 1 h. LC-MS showed the reaction was completed, and one main peak with the desired m/z was detected. The reaction mixture was

quenched by the addition of H2O 10 mL, and extracted with DCM 10 mL (5 mL × 2). The combined organic layers were washed with brine 10 mL, dried over Na2SO4, filtered and concentrated under

reduced pressure to give a residue, which was washed by MeCN (10 mL), then filtered to collect the solid. Compound 3-[(2-chloroacetyl)amino]-N-(3-methoxyphenyl) benzamide (50.49 mg, 155.04

µmol, 37.56% yield, 97.879% purity) was obtained as an off-white solid. ESI-MS m/z: calcd for C16H15ClN2O3 318.08, m/z found 319.1 [M + H]+. 1H-NMR (DMSO, 400 MHz): δ ppm 10.50 (br s, 1H),

10.25 (br s, 1H), 8.09 (br s, 1H), 7.82 (br d, _J_ = 8.4 Hz, 1H), 7.66 (br d, _J_ = 7.2 Hz, 1H), 7.54–7.42 (m, 2H), 7.36 (br d, _J_ = 8.0 Hz, 1H), 7.30–7.21 (m, 1H), 6.69 (br d, _J_ = 8.0 

Hz, 1H), 4.28 (s, 2H), 3.75 (s, 3H). CTPS1 INHIBITOR TREATMENT AND CELL PROLIFERATION ASSAY KMS-11, KMS-12 and NCI-H929 cell lines were treated with 2 µM of the _CTPS1_ inhibitor for 24, 48,

72 and 96 h. After the indicated times of treatment, cell proliferation was analyzed using the CellTiter 96 Aqueous One Solution Cell Proliferation Assay (Promega, Madison, W) following the

manufacturer’s instructions. This is a colorimetric method for determining the number of viable cells in proliferation. For the assay, 100,000 cells by triplicate were plated in 96 wells

plates after _CTPS1_ inhibitor treatment. Plates were centrifuged at 800×_g_ for 10 min, and the medium was removed. Then, cells were incubated with 100 μL per well of medium and 20 μL per

well of CellTiter 96 Aqueous One Solution reagent. After 1–3 h of incubation at 37 °C, the plates were incubated for 1–4 h, depending on the cell line, at 37 °C in a humidified, 5% CO2

atmosphere. The absorbance was recorded at 490 nm and at 650 nm as a reference wavelength using 96-well plate readers until the absorbance of control cells without treatment was around 0.8.

The background absorbance was measured in wells with only cell line medium and solution reagent. First, the average of the absorbance from the control wells was subtracted from all other

absorbance values. Data were calculated as the percentage of total absorbance of treated cells/absorbance of non-treated cells. CRISPR-CAS9 CTPS1 KNOCKOUT To perform CRISPR-Cas9 experiments,

we first generated three stable Cas9 cell lines: KMS-11, RPMI-8226 and KMS-12, infecting them with lentiviruses carrying the Cas9-2A-Blasticidin expressing cassette (Lenti-Cas9-2A-Blast

plasmid, Addgene #73310) following standard protocol, described before, and selecting them with 17.5 μg/mL blasticidin (Invitrogen) for 7 days. For _CTPS1_ knockout, we designed three

different gRNAs (sgCTPS1-1 F: GCTGATGAAATGGAAAGAGA, sgCTPS1-1 R: TCTCTTTCCATTTCATCAGC; sgCTPS1-3 F: CCTCGAACACCAAATCCTCC, sgCTPS1-3 R: GGAGGATTTGGTGTTCGAGG; sgCTPS1-4 F:

ACTTGGGCTTTCCCCAGATC, sgCTPS1-4 R: GATCTGGGGAAAGCCCAAGT), and we cloned them into CRISPseq-BFP-backbone (Addgene #85707) harboring a BFP reporter gene. Corresponding lentiviruses were

produced for all the constructs and constitutively Cas9 expressing cell lines were infected. BFP+ cells in the case of single gRNAs were determined by flow cytometry (FACS Canto II, BD

Biosciences) 72 h after infection. As a control, cells infected with empty CRISPseq-BFP-backbone were used (scramble, Scr). To assess efficiency in the _CTPS1_ knockout experiment, we

checked _CTPS1_ protein level by western blot 7 days after lentiviruses infection. The detailed gating strategy for flow cytometry is shown in Supplementary Fig. 20a. WESTERN BLOT Cells

protein extracts were obtained by lysis buffer containing 1% Triton X-100, 150 mM NaCl, 50 mM Tris [pH 8], supplemented with 1x protease inhibitor cocktail (Complete Mini, Roche), 10 mM NaF

and 1 mM sodium orthovanadate for 30 min at 4 °C and quantified using BCA (Thermo Fisher Scientific) colorimetric assays. A 30-µg amount of protein was separated on 10% sodium dodecyl

sulfate-polyacrylamide gel electrophoresis and transferred to a nitrocellulose membrane. The membrane, after being blocked with Tropix I-block blocking reagent (Cat No AI300, Tropix) in PBS

with 0.1% Tween 20 and 0.02 NaN3 was incubated with the primary antibody against _CTPS1_ (rabbit polyclonal antibody, Cat No 98287, Cell Signaling) diluted 1:1000 o/n at 4 °C. Bound

antibodies were revealed by a chemiluminescent reagent (Tropix) and visualized using the Chemidoc Imaging Systems (Bio-Rad Laboratories). β-actin was used as a loading control (diluted

1:4000 o/n at 4 °C or for 1 h at room temperature) (Anti-β-actin, A5441; Sigma). CELLS TRANSFECTION Cells were passaged 24 h before nucleofection. The transfection of siRNAs was done with

the Nucleofector II device (Amaxa GmbH, Köln, Germany) following the Amaxa guidelines. Briefly, 1 × 106 of JJN-3 and OPM-2 were resuspended in 100 µL of supplemented culture medium with 100 

nM of _UAP1_ siRNAs or Silencer Select Negative Control-1 siRNA (Ambion, Austin, TX) and nucleofected with the Amaxa nucleofector apparatus using programs A-033, A-020 and G-016,

respectively. We used two different siRNAs against _UAP1_ target (siUPA1 A: GUACUUUGGUUUAAAAAAA; siUPA1 B: AGAGCAUUAUGAACAUUAA) to demonstrate that the results obtained with _UAP1_ siRNA

nucleofection are not due to a combination of inconsistent silencing and sequence-specific off-target effects. Silencer Select Negative Control-1 siRNA was used to demonstrate that the

nucleofection did not induce non-specific effects on gene expression. Nucleofection was performed twice with a 24 h interval. After 48 h of the second nucleofection, the _UAP1_ mRNA

expression was analyzed by q-PCR (GUS was employed as the reference gene). Cell proliferation was analyzed 0, 2, 4 and 6 days after two repetitive transfections using the CellTiter 96

Aqueous One Solution Cell Proliferation Assay (Promega, Madison, W) as described above. Transfection efficiency was determined by flow cytometry using the BLOCK IT Fluorescent Oligo

(Invitrogen Life Technologies, Paisley, UK). The detailed gating strategy for flow cytometry is shown in Supplementary Fig. 20b. STATISTICS AND REPRODUCIBILITY Statistical analyses were

performed in R (v4.0.5). Each statistical method used is specified in the “Results” section, including correlation tests for DepMap Analysis, the hypergeometric test for assessing the

association of gene paralogous pairs with synthetic lethality, unpaired one-sided Wilcoxon tests for comparing mean differences of variables without normal distribution and Fisher’s test for

the validation of predicted lethal DKOs with paralogous genes. The number of human samples or cells analyzed in each figure is indicated in their associated legends. For in vitro

experiments, no statistical method was used to predetermine sample size, and no data were excluded from the analyses performed. All experiments and measures were made in triplicate obtaining

in all of the cases similar results. IMPLEMENTATION AND AVAILABILITY gmctool has been developed using R47 and Shiny48. gmctool is hosted using the Amazon Web Services cloud environment

service, and it can be publicly accessed at: https://biotecnun.unav.es/app/gmctool. A full tutorial and example of our own cohort of samples corresponding to B-cell differentiation and MM

can be found on the ‘Help’ tab of gmctool. REPORTING SUMMARY Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. DATA

AVAILABILITY The authors declare that all data supporting the findings of this study are available within the article, in the supplementary material or in other studies. In addition, the

source data of Figs. 1d, a, b, 4c–f, 5c–e, Supplementary Figs. 4, 5, 7c, 8c, 9c, 10c, 11c, 14, 15, 16c, 17c, 18c and 19 are provided as a Source Data file. Raw data of Fig. 2b, c can be

found in Supplementary Data 1. Referenced accession: B-cell and MM RNA-seq data was obtained from GEO under accession codes GSE15106322,23

[https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE101063] and GSE11481621 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE114816]. MMRF-CoMMpass data were generated as part of

the Multiple Myeloma Research Foundation Personalized Medicine Initiatives (https://research.themmrf.org and www.themmrf.org). Cancer cell line data, including RNA-seq, proteomics and

large-scale CRISPR genetic screens, was obtained from www.depmap.org11,12 (release 21Q2). Raw data of Figs. 1c, 3c, 4a, b, 5a, b, Supplementary Figs. 7a, b, 8a, b, 9a, b, 10a, b, 11a, b,

16a, b, 17a, b and 18a, b can be found in the referenced accessions detailed above. Exact _p_-values found in panel b of Figs. 4, 5, Supplementary Figs. 7–11, 16–18 can be found in the

Source Data file. Source data are provided with this paper. CODE AVAILABILITY The code for gmctool is available on GitHub (https://github.com/lvalcarcel/gmctool) and Zenodo

(https://doi.org/10.5281/zenodo.13285245)49. Using this tool, it is possible to generate all the results presented in this article. Gene expression data should be a matrix which has genes

(in ENSEMBL annotation) for rows and samples for columns. REFERENCES * Robinson, J. L. et al. An atlas of human metabolism. _Sci. Signal._ 13, 1–12 (2020). Article  Google Scholar  * Cook,

D. J. & Nielsen, J. Genome-scale metabolic models applied to human health and disease. _Wiley Interdiscip. Rev. Syst. Biol. Med._ 9, e1393 (2017). Article  Google Scholar  * Lewis, N. E.

& Abdel-Haleem, A. M. The evolution of genome-scale models of cancer metabolism. _Front. Physiol._ 0, 237 (2013). Google Scholar  * Ghaffari, P., Mardinoglu, A. & Nielsen, J. Cancer

metabolism: a modeling perspective. _Front. Physiol._ 0, 382 (2015). Google Scholar  * Lewis, N. E., Nagarajan, H. & Palsson, B. O. Constraining the metabolic genotype–phenotype

relationship using a phylogeny of in silico methods. _Nat. Rev. Microbiol._ 10, 291–305 (2012). Article  CAS  PubMed  PubMed Central  Google Scholar  * Tobalina, L., Pey, J., Rezola, A.

& Planes, F. J. Assessment of FBA based gene essentiality analysis in cancer with a fast context-specific network reconstruction method. _PLoS ONE_ 11, e0154583 (2016). Article  PubMed 

PubMed Central  Google Scholar  * Folger, O. et al. Predicting selective drug targets in cancer through metabolic networks. _Mol. Syst. Biol._ 7, 501 (2011). Article  CAS  PubMed  PubMed

Central  Google Scholar  * Apaolaza, I. et al. An in-silico approach to predict and exploit synthetic lethality in cancer metabolism. _Nat. Commun._ 8, 1–9 (2017). Article  CAS  Google

Scholar  * Apaolaza, I., Valcarcel, L. V. & Planes, F. J. GMCS: fast computation of genetic minimal cut sets in large networks. _Bioinformatics_ 35, 535–537 (2019). Article  CAS  PubMed

  Google Scholar  * Thiele, I. et al. A community-driven global reconstruction of human metabolism. _Nat. Biotechnol._ 31, 419–425 (2013). Article  CAS  PubMed  Google Scholar  * Meyers, R.

M. et al. Computational correction of copy number effect improves specificity of CRISPR-Cas9 essentiality screens in cancer cells. _Nat. Genet._ 49, 1779–1784 (2017). Article  CAS  PubMed 

PubMed Central  Google Scholar  * Ghandi, M. et al. Next-generation characterization of the Cancer Cell Line Encyclopedia. _Nature_ 569, 503–508 (2019). Article  ADS  CAS  PubMed  PubMed

Central  Google Scholar  * Agren, R. et al. Identification of anticancer drugs for hepatocellular carcinoma through personalized genome-scale metabolic modeling. _Mol. Syst. Biol._ 10, 721

(2014). Article  PubMed  PubMed Central  Google Scholar  * Costanzo, M. et al. A global genetic interaction network maps a wiring diagram of cellular function. _Science_ 353, aaf1420 (2016).

Article  PubMed  PubMed Central  Google Scholar  * McCall, M. N. et al. The Gene Expression Barcode 3.0: improved data processing and mining tools. _Nucleic Acids Res._ 42, D938–D943

(2014). Article  CAS  PubMed  Google Scholar  * Richelle, A., Joshi, C. & Lewis, N. E. Assessing key decisions for transcriptomic data integration in biochemical networks. _PLoS Comput.

Biol._ 15, e1007185 (2019). Article  ADS  PubMed  PubMed Central  Google Scholar  * Gonatopoulos-Pournatzis, T. et al. Genetic interaction mapping and exon-resolution functional genomics

with a hybrid Cas9-Cas12a platform. _Nat. Biotechnol._ 38, 638–648 (2020). Article  CAS  PubMed  Google Scholar  * Thompson, N. A. et al. Combinatorial CRISPR screen identifies fitness

effects of gene paralogues. _Nat. Commun._ 12, 1302 (2021). Article  ADS  CAS  PubMed  PubMed Central  Google Scholar  * De Kegel, B., Quinn, N., Thompson, N. A., Adams, D. J. & Ryan, C.

J. Comprehensive prediction of robust synthetic lethality between paralog pairs in cancer cell lines. _Cell Syst._ 12, 1144–1159.e6 (2021). Article  PubMed  Google Scholar  * Muller, F. L.

et al. Passenger deletions generate therapeutic vulnerabilities in cancer. _Nature_ 488, 337–342 (2012). Article  ADS  CAS  PubMed  PubMed Central  Google Scholar  * Agirre, X. et al. Long

non-coding RNAs discriminate the stages and gene regulatory states of human humoral immune response. _Nat. Commun._ 10, 821 (2019). Article  ADS  CAS  PubMed  PubMed Central  Google Scholar

  * Ordoñez, R. et al. Chromatin activation as a unifying principle underlying pathogenic mechanisms in multiple myeloma. _Genome Res._ 30, 1217–1227 (2020). Article  PubMed  PubMed Central

  Google Scholar  * Carrasco-Leon, A. et al. Characterization of complete lncRNAs transcriptome reveals the functional and clinical impact of lncRNAs in multiple myeloma. _Leukemia_ 35,

1438–1450 (2021). Article  CAS  PubMed  PubMed Central  Google Scholar  * Martin, E. et al. CTP synthase 1 deficiency in humans reveals its central role in lymphocyte proliferation. _Nature_

510, 288–292 (2014). Article  ADS  CAS  PubMed  PubMed Central  Google Scholar  * Lynch, E. M. et al. Structural basis for isoform-specific inhibition of human CTPS1. _Proc. Natl Acad. Sci.

USA_ 118, e2107968118 (2021). Article  CAS  PubMed  PubMed Central  Google Scholar  * Chiron, D. et al. STP938, a selective CTPS1 inhibitor, shows single agent activity and synergy with

BCL2 inhibition in preclinical models of mantle cell lymphoma. _Blood_ 140, 6014–6015 (2022). Article  Google Scholar  * Rao, Y. et al. Targeting CTP synthetase 1 to restore interferon

induction and impede nucleotide synthesis in SARS-CoV-2 infection. Preprint at _bioRxiv_ https://doi.org/10.1101/2021.02.05.429959 (2021). * Wu, J. L. et al. O-GlcNAcylation is required for

B cell homeostasis and antibody responses. _Nat. Commun._ 8, 1–10 (2017). Article  ADS  Google Scholar  * Chesi, M. et al. Frequent translocation t(4;14)(p16.3;q32.3) in multiple myeloma is

associated with increased expression and activating mutations of fibroblast growth factor receptor 3. _Nat. Genet._ 16, 260–264 (1997). Article  CAS  PubMed  PubMed Central  Google Scholar 

* Colaprico, A. et al. TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGA data. _Nucleic Acids Res._ 44, e71–e71 (2016). Article  PubMed  Google Scholar  * Piao, C. et

al. Inhibition of stearoyl CoA desaturase-1 activity suppresses tumour progression and improves prognosis in human bladder cancer. _J. Cell. Mol. Med._ 23, 2064–2076 (2019). Article  CAS 

PubMed  Google Scholar  * Chajès, V., Cambot, M., Moreau, K., Lenoir, G. M. & Joulin, V. Acetyl-CoA carboxylase α is essential to breast cancer cell survival. _Cancer Res._ 66, 5287–5294

(2006). Article  PubMed  Google Scholar  * Li, S. et al. Acetyl-CoA-carboxylase 1-mediated de novo fatty acid synthesis sustains Lgr5+ intestinal stem cell function. _Nat. Commun._ 13, 3998

(2022). Article  ADS  CAS  PubMed  PubMed Central  Google Scholar  * Agren, R. et al. Reconstruction of genome-scale active metabolic networks for 69 human cell types and 16 cancer types

using INIT. _PLoS Comput. Biol._ 8, e1002518 (2012). Article  CAS  PubMed  PubMed Central  Google Scholar  * Dikicioglu, D., Kırdar, B. & Oliver, S. G. Biomass composition: the “elephant

in the room” of metabolic modelling. _Metabolomics_ 11, 1690–1701 (2015). Article  CAS  PubMed  PubMed Central  Google Scholar  * Lachance, J. C. et al. BOFdat: generating biomass objective

functions for genome-scale metabolic models from experimental data. _PLoS Comput. Biol._ 15, e1006971 (2019). Article  PubMed  PubMed Central  Google Scholar  * Moscardó García, M.,

Pacheco, M., Bintener, T., Presta, L. & Sauter, T. Importance of the biomass formulation for cancer metabolic modeling and drug prediction. _iScience_ 24, 103110 (2021). Article  ADS 

PubMed  PubMed Central  Google Scholar  * Hernández Patiño, C. E., Jaime-Muñoz, G. & Resendis-Antonio, O. Systems biology of cancer: moving toward the integrative study of the metabolic

alterations in cancer cells. _Front. Physiol._ 3, 481 (2013). Article  PubMed  PubMed Central  Google Scholar  * Zielinski, D. C. et al. Systems biology analysis of drivers underlying

hallmarks of cancer cell metabolism. _Sci. Rep._ 7, 1–14 (2017). Article  CAS  Google Scholar  * Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. _Cell_ 144,

646–674 (2011). Article  CAS  PubMed  Google Scholar  * Pey, J. et al. In-silico gene essentiality analysis of polyamine biosynthesis reveals APRT as a potential target in cancer. _Sci.

Rep._ 7, 1–10 (2017). Article  CAS  Google Scholar  * Duarte, N. C. et al. Global reconstruction of the human metabolic network based on genomic and bibliomic data. _Proc. Natl Acad. Sci.

USA_ 104, 1777–1782 (2007). Article  ADS  CAS  PubMed  PubMed Central  Google Scholar  * Swainston, N. et al. Recon 2.2: from reconstruction to model of human metabolism. _Metabolomics_ 12,

1–7 (2016). Article  CAS  Google Scholar  * Orth, J. D., Thiele, I. & Palsson, B. Ø. What is flux balance analysis? _Nat. Biotechnol._ 28, 245–248 (2010). Article  CAS  PubMed  PubMed

Central  Google Scholar  * Heirendt, L. et al. Creation and analysis of biochemical constraint-based models using the COBRA Toolbox v.3.0. _Nat. Protoc._ 14, 639–702 (2019). Article  CAS 

PubMed  PubMed Central  Google Scholar  * Wang, H. et al. RAVEN 2.0: a versatile toolbox for metabolic network reconstruction and a case study on _Streptomyces coelicolor_. _PLoS Comput.

Biol._ 14, 1–17 (2018). Article  ADS  Google Scholar  * R Core Team. R: a language and environment for statistical computing (R Foundation for Statistical Computing, Vienna, Austria, 2021).

https://www.R-project.org/. * Chang, W. et al. shiny: web application framework for R (2021). https://shiny.rstudio.com/, https://shiny.posit.co * Valcarcel, L. V. et al. Scripts and code

for ‘gmctool: automated network-based tool to search for metabolic vulnerabilities in cancer’. _Zenodo_ https://doi.org/10.5281/zenodo.13285245 (2024). Download references ACKNOWLEDGEMENTS

This work was supported by the Minister of Economy and Competitiveness of Spain [PID2019-110344RB-I00 and PID2022-143298OB-I00], PIBA Programme of the Basque Government [PIBA_2020_01_0055],

Elkartek programme of the Basque Government [KK-2020/00008], Fundación Ramon Areces [PREMMAM] to F.J.P.; Instituto de Salud Carlos III (ISCIII) [PI16/02024, PI17/00701, PI19/01352,

PI20/01306], CIBERONC (Co-financed with European Union FEDER funds) [CB16/12/00489], ERANET program ERAPerMed [MEET-AML], MINECO Explora [RTHALMY], Cancer Research UK and AECC under the

Accelerator Award Programme [C355/A26819]. to F.P., Instituto de Salud Carlos III (ISCIII) [FI17/00297] to L.V.V. and Basque Government predoctoral grant [PRE_2021_2_0025] to N.B. AUTHOR

INFORMATION AUTHORS AND AFFILIATIONS * University of Navarra, Tecnun School of Engineering, Manuel de Lardizábal 13, 20018, San Sebastián, Spain Luis V. Valcárcel, Iñigo Apaolaza, Danel

Olaverri-Mendizabal, Naroa Barrena & Francisco J. Planes * Hemato-Oncology Program, Center for Applied Medical Research (CIMA), Universidad de Navarra, IDISNA, CCUN, Avenida Pío XII 55,

31008, Pamplona, Spain Luis V. Valcárcel, Edurne San José-Enériz, Raquel Ordoñez, Ana Valcárcel, Leire Garate, Jesús San Miguel, Antonio Pineda-Lucena, Xabier Agirre & Felipe Prósper *

CIBERONC Centro de Investigación Biomédica en Red de Cáncer, 28029, Madrid, Spain Edurne San José-Enériz, Raquel Ordoñez, Leire Garate, Jesús San Miguel, Antonio Pineda-Lucena, Xabier Agirre

 & Felipe Prósper * Departmento de Hematología, Clínica Universidad de Navarra and CCUN, Universidad de Navarra, Avenida Pío XII 36, 31008, Pamplona, Spain Jesús San Miguel & Felipe

Prósper * Biomedical Engineering Center, University of Navarra, 31008, Pamplona, Navarra, Spain Francisco J. Planes * University of Navarra, Instituto de Ciencia de los Datos e Inteligencia

Artificial (DATAI), Campus Universitario, 31008, Pamplona, Spain Francisco J. Planes Authors * Luis V. Valcárcel View author publications You can also search for this author inPubMed Google

Scholar * Edurne San José-Enériz View author publications You can also search for this author inPubMed Google Scholar * Raquel Ordoñez View author publications You can also search for this

author inPubMed Google Scholar * Iñigo Apaolaza View author publications You can also search for this author inPubMed Google Scholar * Danel Olaverri-Mendizabal View author publications You

can also search for this author inPubMed Google Scholar * Naroa Barrena View author publications You can also search for this author inPubMed Google Scholar * Ana Valcárcel View author

publications You can also search for this author inPubMed Google Scholar * Leire Garate View author publications You can also search for this author inPubMed Google Scholar * Jesús San

Miguel View author publications You can also search for this author inPubMed Google Scholar * Antonio Pineda-Lucena View author publications You can also search for this author inPubMed 

Google Scholar * Xabier Agirre View author publications You can also search for this author inPubMed Google Scholar * Felipe Prósper View author publications You can also search for this

author inPubMed Google Scholar * Francisco J. Planes View author publications You can also search for this author inPubMed Google Scholar CONTRIBUTIONS X.A., F.P., and F.J.P. conceived this

study. L.V.V., I.A. and F.J.P. developed the conceptual approach, and N.B., D.O. and L.V.V. performed the computational analysis. E.S.J.-E., R.O., A.V., L.G., A.P.-L. and X.A. performed the

experiments. F.P., X.A., E.S.J.-E. and J.S.-M. analyzed and provided access to multiple myeloma data. All authors wrote, read, and approved the manuscript. CORRESPONDING AUTHORS

Correspondence to Felipe Prósper or Francisco J. Planes. ETHICS DECLARATIONS COMPETING INTERESTS The authors declare no competing interests. PEER REVIEW PEER REVIEW INFORMATION _Nature

Communications_ thanks Niels Weinhold and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available. ADDITIONAL INFORMATION

PUBLISHER’S NOTE Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. SUPPLEMENTARY INFORMATION SUPPLEMENTARY INFORMATION

PEER REVIEW FILE DESCRIPTION OF ADDITIONAL SUPPLEMENTARY INFORMATION SUPPLEMENTARY DATA 1 SUPPLEMENTARY DATA 2 SUPPLEMENTARY DATA 3 SUPPLEMENTARY DATA 4 SUPPLEMENTARY DATA 5 SUPPLEMENTARY

DATA 6 SUPPLEMENTARY DATA 7 REPORTING SUMMARY SOURCE DATA SOURCE DATA RIGHTS AND PERMISSIONS OPEN ACCESS This article is licensed under a Creative Commons

Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give

appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission

under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons

licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by

statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit

http://creativecommons.org/licenses/by-nc-nd/4.0/. Reprints and permissions ABOUT THIS ARTICLE CITE THIS ARTICLE Valcárcel, L.V., San José-Enériz, E., Ordoñez, R. _et al._ An automated

network-based tool to search for metabolic vulnerabilities in cancer. _Nat Commun_ 15, 8685 (2024). https://doi.org/10.1038/s41467-024-52725-4 Download citation * Received: 22 June 2022 *

Accepted: 18 September 2024 * Published: 11 October 2024 * DOI: https://doi.org/10.1038/s41467-024-52725-4 SHARE THIS ARTICLE Anyone you share the following link with will be able to read

this content: Get shareable link Sorry, a shareable link is not currently available for this article. Copy to clipboard Provided by the Springer Nature SharedIt content-sharing initiative