An expanded chemical space is essential for improved identification of small molecules for emerging therapeutic targets. However, the identification of targets for novel compounds is biased towards the synthesis of known scaffolds that bind familiar protein families, limiting the exploration of chemical space. To change this paradigm, we validated a new pipeline that identifies small molecule-protein interactions and works even for compounds lacking similarity to known drugs. Based on differential mRNA profiles in multiple cell types exposed to drugs and in which gene knockdowns (KD) were conducted, we showed that drugs induce gene regulatory networks that correlate with those produced after silencing protein-coding genes. Next, we applied supervised machine learning to exploit drug-KD signature correlations and enriched our predictions using an orthogonal structure-based screen. As a proof-of-principle for this regimen, top-10/top-100 target prediction accuracies of 26% and 41%, respectively, were achieved on a validation of set 152 FDA-approved drugs and 3104 potential targets. We then predicted targets for 1680 compounds and validated chemical interactors with four targets that have proven difficult to chemically modulate, including non-covalent inhibitors of HRAS and KRAS. Importantly, drug-target interactions manifest as gene expression correlations between drug treatment and both target gene KD and KD of genes that act up- or down-stream of the target, even for relatively weak binders. These correlations provide new insights on the cellular response of disrupting protein interactions and highlight the complex genetic phenotypes of drug treatment. With further refinement, our pipeline may accelerate the identification and development of novel chemical classes by screening compound-target interactions.

Transcriptional regulation of some genes involved in xenobiotic detoxification and apoptosis is performed via the human pregnane X receptor (PXR) which in turn is activated by structurally diverse agonists including steroid hormones. Activation of PXR has the potential to initiate adverse effects, altering drug pharmacokinetics or perturbing physiological processes. Reliable computational prediction of PXR agonists would be valuable for pharmaceutical and toxicological research. There has been limited success with structure-based modeling approaches to predict human PXR activators. Slightly better success has been achieved with ligand-based modeling methods including quantitative structure-activity relationship (QSAR) analysis, pharmacophore modeling and machine learning. In this study, we present a comprehensive analysis focused on prediction of 115 steroids for ligand binding activity towards human PXR. Six crystal structures were used as templates for docking and ligand-based modeling approaches (two-, three-, four- and five-dimensional analyses). The best success at external prediction was achieved with 5D-QSAR. Bayesian models with FCFP_6 descriptors were validated after leaving a large percentage of the dataset out and using an external test set. Docking of ligands to the PXR structure co-crystallized with hyperforin had the best statistics for this method. Sulfated steroids (which are activators) were consistently predicted as non-activators while, poorly predicted steroids were docked in a reverse mode compared to 5alpha-androstan-3beta-ol. Modeling of human PXR represents a complex challenge by virtue of the large, flexible ligand-binding cavity. This study emphasizes this aspect, illustrating modest success using the largest quantitative data set to date and multiple modeling approaches.

Computational methods play a pivotal role in drug discovery and are widely applied in virtual screening, structure optimization, and compound activity profiling. Over the last decades, almost all the attention in medicinal chemistry has been directed to protein-ligand binding, and computational tools have been created with this target in mind. With novel discoveries of functional RNAs and their possible applications, RNAs have gained considerable attention as potential drug targets. However, the availability of bioinformatics tools for nucleic acids is limited. Here, we introduce fingeRNAt-a software tool for detecting non-covalent interactions formed in complexes of nucleic acids with ligands. The program detects nine types of interactions: (i) hydrogen and (ii) halogen bonds, (iii) cation-anion, (iv) pi-cation, (v) pi-anion, (vi) pi-stacking, (vii) inorganic ion-mediated, (viii) water-mediated, and (ix) lipophilic interactions. However, the scope of detected interactions can be easily expanded using a simple plugin system. In addition, detected interactions can be visualized using the associated PyMOL plugin, which facilitates the analysis of medium-throughput molecular complexes. Interactions are also encoded and stored as a bioinformatics-friendly Structural Interaction Fingerprint (SIFt)-a binary string where the respective bit in the fingerprint is set to 1 if a particular interaction is present and to 0 otherwise. This output format, in turn, enables high-throughput analysis of interaction data using data analysis techniques. We present applications of fingeRNAt-generated interaction fingerprints for visual and computational analysis of RNA-ligand complexes, including analysis of interactions formed in experimentally determined RNA-small molecule ligand complexes deposited in the Protein Data Bank. We propose interaction fingerprint-based similarity as an alternative measure to RMSD to recapitulate complexes with similar interactions but different folding. We present an application of interaction fingerprints for the clustering of molecular complexes. This approach can be used to group ligands that form similar binding networks and thus have similar biological properties. The fingeRNAt software is freely available at https://github.com/n-szulc/fingeRNAt.

Calmodulin (CaM) is a calcium sensor which binds and regulates a wide range of target-proteins. This implicitly enables the concentration of calcium to influence many downstream physiological responses, including muscle contraction, learning and depression. The antipsychotic drug trifluoperazine (TFP) is a known CaM inhibitor. By binding to various sites, TFP prevents CaM from associating to target-proteins. However, the molecular and state-dependent mechanisms behind CaM inhibition by drugs such as TFP are largely unknown. Here, we build a Markov state model (MSM) from adaptively sampled molecular dynamics simulations and reveal the structural and dynamical features behind the inhibitory mechanism of TFP-binding to the C-terminal domain of CaM. We specifically identify three major TFP binding-modes from the MSM macrostates, and distinguish their effect on CaM conformation by using a systematic analysis protocol based on biophysical descriptors and tools from machine learning. The results show that depending on the binding orientation, TFP effectively stabilizes features of the calcium-unbound CaM, either affecting the CaM hydrophobic binding pocket, the calcium binding sites or the secondary structure content in the bound domain. The conclusions drawn from this work may in the future serve to formulate a complete model of pharmacological modulation of CaM, which furthers our understanding of how these drugs affect signaling pathways as well as associated diseases.

Fungal infection has become one of the leading causes of hospital-acquired infections with high mortality rates. Furthermore, drug resistance is common for fungus-causing diseases. Synergistic drug combinations could provide an effective strategy to overcome drug resistance. Meanwhile, synergistic drug combinations can increase treatment efficacy and decrease drug dosage to avoid toxicity. Therefore, computational prediction of synergistic drug combinations for fungus-causing diseases becomes attractive. In this study, we proposed similar nature of drug combinations: principal drugs which obtain synergistic effect with similar adjuvant drugs are often similar and vice versa. Furthermore, we developed a novel algorithm termed Network-based Laplacian regularized Least Square Synergistic drug combination prediction (NLLSS) to predict potential synergistic drug combinations by integrating different kinds of information such as known synergistic drug combinations, drug-target interactions, and drug chemical structures. We applied NLLSS to predict antifungal synergistic drug combinations and showed that it achieved excellent performance both in terms of cross validation and independent prediction. Finally, we performed biological experiments for fungal pathogen Candida albicans to confirm 7 out of 13 predicted antifungal synergistic drug combinations. NLLSS provides an efficient strategy to identify potential synergistic antifungal combinations.

In pharmaceutical sciences, a crucial step of the drug discovery process is the identification of drug-target interactions. However, only a small portion of the drug-target interactions have been experimentally validated, as the experimental validation is laborious and costly. To improve the drug discovery efficiency, there is a great need for the development of accurate computational approaches that can predict potential drug-target interactions to direct the experimental verification. In this paper, we propose a novel drug-target interaction prediction algorithm, namely neighborhood regularized logistic matrix factorization (NRLMF). Specifically, the proposed NRLMF method focuses on modeling the probability that a drug would interact with a target by logistic matrix factorization, where the properties of drugs and targets are represented by drug-specific and target-specific latent vectors, respectively. Moreover, NRLMF assigns higher importance levels to positive observations (i.e., the observed interacting drug-target pairs) than negative observations (i.e., the unknown pairs). Because the positive observations are already experimentally verified, they are usually more trustworthy. Furthermore, the local structure of the drug-target interaction data has also been exploited via neighborhood regularization to achieve better prediction accuracy. We conducted extensive experiments over four benchmark datasets, and NRLMF demonstrated its effectiveness compared with five state-of-the-art approaches.

Elucidating the causal mechanisms responsible for disease can reveal potential therapeutic targets for pharmacological intervention and, accordingly, guide drug repositioning and discovery. In essence, the topology of a network can reveal the impact a drug candidate may have on a given biological state, leading the way for enhanced disease characterization and the design of advanced therapies. Network-based approaches, in particular, are highly suited for these purposes as they hold the capacity to identify the molecular mechanisms underlying disease. Here, we present drug2ways, a novel methodology that leverages multimodal causal networks for predicting drug candidates. Drug2ways implements an efficient algorithm which reasons over causal paths in large-scale biological networks to propose drug candidates for a given disease. We validate our approach using clinical trial information and demonstrate how drug2ways can be used for multiple applications to identify: i) single-target drug candidates, ii) candidates with polypharmacological properties that can optimize multiple targets, and iii) candidates for combination therapy. Finally, we make drug2ways available to the scientific community as a Python package that enables conducting these applications on multiple standard network formats.

Antibiotics need to be effective in diverse environments in vivo. However, the pathogen microenvironment can have a significant impact on antibiotic potency. Further, antibiotics are increasingly used in combinations to combat resistance, yet, the effect of microenvironments on drug-combination efficacy is unknown. To exhaustively explore the impact of diverse microenvironments on drug-combinations, here we develop a computational framework-Metabolism And GENomics-based Tailoring of Antibiotic regimens (MAGENTA). MAGENTA uses chemogenomic profiles of individual drugs and metabolic perturbations to predict synergistic or antagonistic drug-interactions in different microenvironments. We uncovered antibiotic combinations with robust synergy across nine distinct environments against both E. coli and A. baumannii by searching through 2556 drug-combinations of 72 drugs. MAGENTA also accurately predicted the change in efficacy of bacteriostatic and bactericidal drug-combinations during growth in glycerol media, which we confirmed experimentally in both microbes. Our approach identified genes in glycolysis and glyoxylate pathway as top predictors of synergy and antagonism respectively. Our systems approach enables tailoring of antibiotic therapies based on the pathogen microenvironment.

The third variable loop (V3) of the human immunodeficiency virus type 1 (HIV-1) envelope is a principal determinant of antibody neutralization and progression to AIDS. Although it is undoubtedly an important target for vaccine research, extensive genetic variation in V3 remains an obstacle to the development of an effective vaccine. Comparative methods that exploit the abundance of sequence data can detect interactions between residues of rapidly evolving proteins such as the HIV-1 envelope, revealing biological constraints on their variability. However, previous studies have relied implicitly on two biologically unrealistic assumptions: (1) that founder effects in the evolutionary history of the sequences can be ignored, and; (2) that statistical associations between residues occur exclusively in pairs. We show that comparative methods that neglect the evolutionary history of extant sequences are susceptible to a high rate of false positives (20%-40%). Therefore, we propose a new method to detect interactions that relaxes both of these assumptions. First, we reconstruct the evolutionary history of extant sequences by maximum likelihood, shifting focus from extant sequence variation to the underlying substitution events. Second, we analyze the joint distribution of substitution events among positions in the sequence as a Bayesian graphical model, in which each branch in the phylogeny is a unit of observation. We perform extensive validation of our models using both simulations and a control case of known interactions in HIV-1 protease, and apply this method to detect interactions within V3 from a sample of 1,154 HIV-1 envelope sequences. Our method greatly reduces the number of false positives due to founder effects, while capturing several higher-order interactions among V3 residues. By mapping these interactions to a structural model of the V3 loop, we find that the loop is stratified into distinct evolutionary clusters. We extend our model to detect interactions between the V3 and C4 domains of the HIV-1 envelope, and account for the uncertainty in mapping substitutions to the tree with a parametric bootstrap.

Drug-induced toxicity is a significant problem in clinical care. A key problem here is a general understanding of the molecular mechanisms accompanying the transition from desired drug effects to adverse events following administration of either therapeutic or toxic doses, in particular within a patient context. Here, a comparative toxicity analysis was performed for fifteen hepatotoxic drugs by evaluating toxic changes reflecting the transition from therapeutic drug responses to toxic reactions at the cellular level. By use of physiologically-based pharmacokinetic modeling, in vitro toxicity data were first contextualized to quantitatively describe time-resolved drug responses within a patient context. Comparatively studying toxic changes across the considered hepatotoxicants allowed the identification of subsets of drugs sharing similar perturbations on key cellular processes, functional classes of genes, and individual genes. The identified subsets of drugs were next analyzed with regard to drug-related characteristics and their physicochemical properties. Toxic changes were finally evaluated to predict both molecular biomarkers and potential drug-drug interactions. The results may facilitate the early diagnosis of adverse drug events in clinical application.

Numerous signaling proteins use multivalent binding to increase the specificity and affinity of their interactions within the cell. Enhancement arises because the effective binding constant for multivalent binding is larger than the binding constants for each individual interaction. We seek to gain both qualitative and quantitative understanding of the multivalent interactions of an adaptor protein, growth factor receptor bound protein-2 (Grb2), containing two SH3 domains interacting with the nucleotide exchange factor son-of-sevenless 1 (Sos1) containing multiple polyproline motifs separated by flexible unstructured regions. Grb2 mediates the recruitment of Sos1 from the cytosol to the plasma membrane where it activates Ras by inducing the exchange of GDP for GTP. First, using a combination of evolutionary information and binding energy calculations, we predict an additional polyproline motif in Sos1 that binds to the SH3 domains of Grb2. This gives rise to a total of five polyproline motifs in Sos1 that are capable of binding to the two SH3 domains of Grb2. Then, using a hybrid method combining molecular dynamics simulations and polymer models, we estimate the enhancement in local concentration of a polyproline motif on Sos1 near an unbound SH3 domain of Grb2 when its other SH3 domain is bound to a different polyproline motif on Sos1. We show that the local concentration of the Sos1 motifs that a Grb2 SH3 domain experiences is approximately 1000 times greater than the cellular concentration of Sos1. Finally, we calculate the intramolecular equilibrium constants for the crosslinking of Grb2 on Sos1 and use thermodynamic modeling to calculate the stoichiometry. With these equilibrium constants, we are able to predict the distribution of complexes that form at physiological concentrations. We believe this is the first systematic analysis that combines sequence, structure, and thermodynamic analyses to determine the stoichiometry of the complexes that are dominant in the cellular environment.

Discovery of efficient anti-cancer drug combinations is a major challenge, since experimental testing of all possible combinations is clearly impossible. Recent efforts to computationally predict drug combination responses retain this experimental search space, as model definitions typically rely on extensive drug perturbation data. We developed a dynamical model representing a cell fate decision network in the AGS gastric cancer cell line, relying on background knowledge extracted from literature and databases. We defined a set of logical equations recapitulating AGS data observed in cells in their baseline proliferative state. Using the modeling software GINsim, model reduction and simulation compression techniques were applied to cope with the vast state space of large logical models and enable simulations of pairwise applications of specific signaling inhibitory chemical substances. Our simulations predicted synergistic growth inhibitory action of five combinations from a total of 21 possible pairs. Four of the predicted synergies were confirmed in AGS cell growth real-time assays, including known effects of combined MEK-AKT or MEK-PI3K inhibitions, along with novel synergistic effects of combined TAK1-AKT or TAK1-PI3K inhibitions. Our strategy reduces the dependence on a priori drug perturbation experimentation for well-characterized signaling networks, by demonstrating that a model predictive of combinatorial drug effects can be inferred from background knowledge on unperturbed and proliferating cancer cells. Our modeling approach can thus contribute to preclinical discovery of efficient anticancer drug combinations, and thereby to development of strategies to tailor treatment to individual cancer patients.

Data integration methods are used to obtain a unified summary of multiple datasets. For multi-modal data, we propose a computational workflow to jointly analyze datasets from cell lines. The workflow comprises a novel probabilistic data integration method, named POPLS-DA, for multi-omics data. The workflow is motivated by a study on synucleinopathies where transcriptomics, proteomics, and drug screening data are measured in affected LUHMES cell lines and controls. The aim is to highlight potentially druggable pathways and genes involved in synucleinopathies. First, POPLS-DA is used to prioritize genes and proteins that best distinguish cases and controls. For these genes, an integrated interaction network is constructed where the drug screen data is incorporated to highlight druggable genes and pathways in the network. Finally, functional enrichment analyses are performed to identify clusters of synaptic and lysosome-related genes and proteins targeted by the protective drugs. POPLS-DA is compared to other single- and multi-omics approaches. We found that HSPA5, a member of the heat shock protein 70 family, was one of the most targeted genes by the validated drugs, in particular by AT1-blockers. HSPA5 and AT1-blockers have been previously linked to α-synuclein pathology and Parkinson's disease, showing the relevance of our findings. Our computational workflow identified new directions for therapeutic targets for synucleinopathies. POPLS-DA provided a larger interpretable gene set than other single- and multi-omic approaches. An implementation based on R and markdown is freely available online.

Advancing age is the dominant risk factor for most of the major killer diseases in developed countries. Hence, ameliorating the effects of ageing may prevent multiple diseases simultaneously. Drugs licensed for human use against specific diseases have proved to be effective in extending lifespan and healthspan in animal models, suggesting that there is scope for drug repurposing in humans. New bioinformatic methods to identify and prioritise potential anti-ageing compounds for humans are therefore of interest. In this study, we first used drug-protein interaction information, to rank 1,147 drugs by their likelihood of targeting ageing-related gene products in humans. Among 19 statistically significant drugs, 6 have already been shown to have pro-longevity properties in animal models (p < 0.001). Using the targets of each drug, we established their association with ageing at multiple levels of biological action including pathways, functions and protein interactions. Finally, combining all the data, we calculated a ranked list of drugs that identified tanespimycin, an inhibitor of HSP-90, as the top-ranked novel anti-ageing candidate. We experimentally validated the pro-longevity effect of tanespimycin through its HSP-90 target in Caenorhabditis elegans.

Drug resistance mutations (DRMs) appear in HIV under treatment pressure. DRMs are commonly transmitted to naive patients. The standard approach to reveal new DRMs is to test for significant frequency differences of mutations between treated and naive patients. However, we then consider each mutation individually and cannot hope to study interactions between several mutations. Here, we aim to leverage the ever-growing quantity of high-quality sequence data and machine learning methods to study such interactions (i.e. epistasis), as well as try to find new DRMs. We trained classifiers to discriminate between Reverse Transcriptase Inhibitor (RTI)-experienced and RTI-naive samples on a large HIV-1 reverse transcriptase (RT) sequence dataset from the UK (n ≈ 55, 000), using all observed mutations as binary representation features. To assess the robustness of our findings, our classifiers were evaluated on independent data sets, both from the UK and Africa. Important representation features for each classifier were then extracted as potential DRMs. To find novel DRMs, we repeated this process by removing either features or samples associated to known DRMs. When keeping all known resistance signal, we detected sufficiently prevalent known DRMs, thus validating the approach. When removing features corresponding to known DRMs, our classifiers retained some prediction accuracy, and six new mutations significantly associated with resistance were identified. These six mutations have a low genetic barrier, are correlated to known DRMs, and are spatially close to either the RT active site or the regulatory binding pocket. When removing both known DRM features and sequences containing at least one known DRM, our classifiers lose all prediction accuracy. These results likely indicate that all mutations directly conferring resistance have been found, and that our newly discovered DRMs are accessory or compensatory mutations. Moreover, apart from the accessory nature of the relationships we found, we did not find any significant signal of further, more subtle epistasis combining several mutations which individually do not seem to confer any resistance.

Accurate prediction of cancer drug response (CDR) is a longstanding challenge in modern oncology that underpins personalized treatment. Current computational methods implement CDR prediction by modeling responses between entire drugs and cell lines, without the consideration that response outcomes may primarily attribute to a few finer-level 'subcomponents', such as privileged substructures of the drug or gene signatures of the cancer cell, thus producing predictions that are hard to explain. Herein, we present SubCDR, a subcomponent-guided deep learning method for interpretable CDR prediction, to recognize the most relevant subcomponents driving response outcomes. Technically, SubCDR is built upon a line of deep neural networks that enables a set of functional subcomponents to be extracted from each drug and cell line profile, and breaks the CDR prediction down to identifying pairwise interactions between subcomponents. Such a subcomponent interaction form can offer a traceable path to explicitly indicate which subcomponents contribute more to the response outcome. We verify the superiority of SubCDR over state-of-the-art CDR prediction methods through extensive computational experiments on the GDSC dataset. Crucially, we found many predicted cases that demonstrate the strength of SubCDR in finding the key subcomponents driving responses and exploiting these subcomponents to discover new therapeutic drugs. These results suggest that SubCDR will be highly useful for biomedical researchers, particularly in anti-cancer drug design.

Prediction of clinical drug response (CDR) of cancer patients, based on their clinical and molecular profiles obtained prior to administration of the drug, can play a significant role in individualized medicine. Machine learning models have the potential to address this issue but training them requires data from a large number of patients treated with each drug, limiting their feasibility. While large databases of drug response and molecular profiles of preclinical in-vitro cancer cell lines (CCLs) exist for many drugs, it is unclear whether preclinical samples can be used to predict CDR of real patients. We designed a systematic approach to evaluate how well different algorithms, trained on gene expression and drug response of CCLs, can predict CDR of patients. Using data from two large databases, we evaluated various linear and non-linear algorithms, some of which utilized information on gene interactions. Then, we developed a new algorithm called TG-LASSO that explicitly integrates information on samples' tissue of origin with gene expression profiles to improve prediction performance. Our results showed that regularized regression methods provide better prediction performance. However, including the network information or common methods of including information on the tissue of origin did not improve the results. On the other hand, TG-LASSO improved the predictions and distinguished resistant and sensitive patients for 7 out of 13 drugs. Additionally, TG-LASSO identified genes associated with the drug response, including known targets and pathways involved in the drugs' mechanism of action. Moreover, genes identified by TG-LASSO for multiple drugs in a tissue were associated with patient survival. In summary, our analysis suggests that preclinical samples can be used to predict CDR of patients and identify biomarkers of drug sensitivity and survival.

Failure to demonstrate efficacy and safety issues are important reasons that drugs do not reach the market. An incomplete understanding of how drugs exert their effects hinders regulatory and pharmaceutical industry projections of a drug's benefits and risks. Signaling pathways mediate drug response and while many signaling molecules have been characterized for their contribution to disease or their role in drug side effects, our knowledge of these pathways is incomplete. To better understand all signaling molecules involved in drug response and the phenotype associations of these molecules, we created a novel method, PathFX, a non-commercial entity, to identify these pathways and drug-related phenotypes. We benchmarked PathFX by identifying drugs' marketed disease indications and reported a sensitivity of 41%, a 2.7-fold improvement over similar approaches. We then used PathFX to strengthen signals for drug-adverse event pairs occurring in the FDA Adverse Event Reporting System (FAERS) and also identified opportunities for drug repurposing for new diseases based on interaction paths that associated a marketed drug to that disease. By discovering molecular interaction pathways, PathFX improved our understanding of drug associations to safety and efficacy phenotypes. The algorithm may provide a new means to improve regulatory and therapeutic development decisions.

The recently proposed concept of molecular connectivity maps enables researchers to integrate experimental measurements of genes, proteins, metabolites, and drug compounds under similar biological conditions. The study of these maps provides opportunities for future toxicogenomics and drug discovery applications. We developed a computational framework to build disease-specific drug-protein connectivity maps. We integrated gene/protein and drug connectivity information based on protein interaction networks and literature mining, without requiring gene expression profile information derived from drug perturbation experiments on disease samples. We described the development and application of this computational framework using Alzheimer's Disease (AD) as a primary example in three steps. First, molecular interaction networks were incorporated to reduce bias and improve relevance of AD seed proteins. Second, PubMed abstracts were used to retrieve enriched drug terms that are indirectly associated with AD through molecular mechanistic studies. Third and lastly, a comprehensive AD connectivity map was created by relating enriched drugs and related proteins in literature. We showed that this molecular connectivity map development approach outperformed both curated drug target databases and conventional information retrieval systems. Our initial explorations of the AD connectivity map yielded a new hypothesis that diltiazem and quinidine may be investigated as candidate drugs for AD treatment. Molecular connectivity maps derived computationally can help study molecular signature differences between different classes of drugs in specific disease contexts. To achieve overall good data coverage and quality, a series of statistical methods have been developed to overcome high levels of data noise in biological networks and literature mining results. Further development of computational molecular connectivity maps to cover major disease areas will likely set up a new model for drug development, in which therapeutic/toxicological profiles of candidate drugs can be checked computationally before costly clinical trials begin.

Cancer cells have genetic alterations that often directly affect intracellular protein signaling processes allowing them to bypass control mechanisms for cell death, growth and division. Cancer drugs targeting these alterations often work initially, but resistance is common. Combinations of targeted drugs may overcome or prevent resistance, but their selection requires context-specific knowledge of signaling pathways including complex interactions such as feedback loops and crosstalk. To infer quantitative pathway models, we collected a rich dataset on a melanoma cell line: Following perturbation with 54 drug combinations, we measured 124 (phospho-)protein levels and phenotypic response (cell growth, apoptosis) in a time series from 10 minutes to 67 hours. From these data, we trained time-resolved mathematical models that capture molecular interactions and the coupling of molecular levels to cellular phenotype, which in turn reveal the main direct or indirect molecular responses to each drug. Systematic model simulations identified novel combinations of drugs predicted to reduce the survival of melanoma cells, with partial experimental verification. This particular application of perturbation biology demonstrates the potential impact of combining time-resolved data with modeling for the discovery of new combinations of cancer drugs.