A Hypergraph-Based Model for Predicting Potential Drug Combinations in Cancer Therapy.

Predicting effective drug combinations for cancer treatment using a graph-based approach

Drug combinations offer a solution to two major issues faced by oncologists, drug resistance and tumour heterogeneity, but identifying the optimum combination of drugs for distinct molecular contexts remains challenging. This is particularly true for cancers where limited targeted therapies are available, for example pancreatic or KRAS-mutant colorectal cancers. We have established a high-throughput, systematic platform for screening large numbers of drug combinations in cell line panels. This includes 650 combinations in 48 colorectal cell lines, over 1,300 combinations in 52 breast cancer cell lines, covering all subtypes of the disease, and 650 combinations in 30 pancreatic cell lines. Combinations are screened in an anchor-library format, with an anchor drug used at two fixed concentrations, combined with the library drug at seven discontinuous concentrations. Drugs were chosen on a per-tissue basis, considering existing standards of care, known genomic features and core tissue pathways. We have prioritised drugs that are currently FDA-approved or are in late-stage clinical trials to ensure maximum clinical relevance of our data. Our screen presents response data for over 100,000 drug combination-cell line pairs, plus single agent drug sensitivities for over 4,800 single agent-cell line pairs. Combinations are classified as effective based on their resultant change in cell viability (ΔEMax) and change in drug sensitivity (ΔIC50). Drug combinations are ranked and prioritised based on potency, number of sensitive cell lines, and clinical relevance, i.e. use of FDA-approved drug(s). Validation of synergistic drug combinations, including using alternative inhibitors of promising target combinations, is ongoing. By screening combinations at this scale, we are able to perform statistically-powerful biomarker analysis, including integration with genomic, transcriptomic and proteomic data, and also clinical classifications such as PAM50 status. This study represents one of the largest drug combination screens performed to date, and focuses on two tissues for which targeted drug therapies are currently limited: pancreatic cancer and colorectal cancer. The third tissue studied, breast, represents a highly heterogenous cancer, meaning that selecting appropriate drug combinations is particularly challenging. We have ranked and prioritised effective drug combinations and identified biomarkers for the molecular contexts in which certain combinations are effective. We are currently validating combinations both in-house and with collaborators. Citation Format: Elizabeth A Coker, Patricia Jaaks, Daniel J Vis, Nanne Aben, Syd Barthorpe, Dieudonne van der Meer, Howard Lightfoot, Lodewyk Wessels, Mathew Garnett, GDSC Screening Team. A high-throughput screen to identify effective and translationally-relevant drug combinations for breast, colorectal, and pancreatic cancers [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference on Molecular Targets and Cancer Therapeutics; 2019 Oct 26-30; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2019;18(12 Suppl):Abstract nr C007. doi:10.1158/1535-7163.TARG-19-C007

Combinational cancer therapies can overcome drug resistance, minimize dosage and toxicity, and enhance therapeutic efficacy. However, identifying effective drug combinations remains a significant challenge. Although machine learning (ML) models trained on high-throughput screening data offer a promising strategy, their clinical translation is often limited by discrepancies between in vitro cell line data and patient outcomes. In this study, we present a novel approach that improves the predictive performance of ML models by incorporating pharmacological data from patient-derived xenograft (PDX) models, which more accurately reflect clinical responses. Leveraging gene expression profiles, drug structure information, and ML models based on the high throughput datasets, we predicted effective drug combinations on proprietary PDX models developed at Certic Oncology. Drug combinations targeting the PI3K and mTOR pathways consistently ranked among top candidates across multiple KRAS G12 mutant models, including those derived from non-small cell lung cancer (NSCLC) and gastric cancer. We experimentally evaluated six PI3K inhibitors and four mTOR inhibitors across 13 KRAS G12 mutant PDX models spanning lung, gastric, pancreatic, and colorectal cancers. These results enabled the selection of seven gene biomarkers for further ML model refinement. Retraining the models with PDX-derived data and the expression profiles of the biomarkers significantly improved the predictive accuracy, achieving a Pearson’s correlation coefficient of 0.63 in forecasting the efficacy of PI3K–mTOR combinations on novel KRAS G12 PDX models. Notably, the combination of everolimus and alpelisib—prioritized by our model—has been independently validated in literature and is currently in clinical trials for solid tumors. Citation Format: Yuan-Hung Chien, Raffaella Pippa, Warren Andrews, Long Do. Machine Learning-Guided Discovery of Drug Combinations Targeting PI3K and mTOR Pathways Using Cell Line and PDX Data [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr A014.

To assess the association between the severity of recent exacerbations and 90-day mortality risk in chronic obstructive pulmonary disease patients (COPD) with acute symptoms, focusing on the impact of the treatment regimen and involving 17 different drug combinations. A longitudinal, retrospective analysis was carried out in 495 hospitalized COPD patients aged 40-75 years. Patients' clinical characteristics were recorded and the effects of drug regimens, administered pre and post hospitalization, comprising various combinations of long-acting muscarinic antagonists (LAMA), long-acting beta agonists (LABA), inhaled corticosteroids (ICS), and antibiotics, were compared. A statistical analysis of the primary outcome, 90-day mortality was used to identify patient attributes best predicting mortality. At discharge, 65% of patients were receiving a 3-drug combination, 33% a 2-drug regimen, and 9% a single-drug therapy. Patients discharged on a 3-drug combination treatment had the lowest 90-day mortality rate (4%) compared to 22% for those treated with single-drug regimens. Multivariate analysis revealed that the risk of death on single-drug therapy was more than 5-fold greater (odds ratio 5.08) than in the case of patients discharged on a multi-drug combination regimen. Patients treated and discharged from hospital on a multi-drug regimen following recent COPD exacerbations had significantly better 90-day survival than patients discharged on monotherapy. The severity of exacerbations and nature of the pharmacotherapy were the main predictors of mortality and were indicative for the importance of disease assessment and multi-drug treatment strategies.

Estimating an individual woman's absolute risk for breast cancer is essential for decision making about screening and preventive recommendations. Although the current standard, the Gail model, is well calibrated in populations, it performs poorly for individuals. Mammographic breast density (BD) may improve the predictive accuracy of the Gail model. Prospective observational cohort of 81,777 women in the San Francisco Mammography Registry presenting for mammography during 1993 through 2002 who had no prior diagnosis of breast cancer. Breast density was rated by clinical radiologists using the Breast Imaging Reporting and Data System classification (almost entirely fat; scattered fibroglandular densities; heterogeneously dense; extremely dense). Breast cancer cases were identified through linkage to Northern California Surveillance Epidemiology End Results (SEER) program. We compared the predictive accuracy of models with Gail risk, breast density, and the combination. All models were adjusted for age and ethnicity. During 5.1 years of follow-up, 955 women were diagnosed with invasive breast cancer. The Gail model had modest predictive accuracy (concordance index (c-index) 0.67; 95% CI 0.65-0.68). Adding breast density to the model increased the predictive accuracy to 0.68 (95% CI .66-.70, p < 0.01 compared with the Gail model alone). The model containing only breast density adjusted for age and ethnicity had predictive accuracy equivalent to the Gail model (c-index 0.67, 95% CI 0.65-0.68). The addition of breast density measured by BI-RADS categories minimally improved the predictive accuracy of the Gail model. A model based on breast density alone adjusted for age and ethnicity was as accurate as the Gail model.

Motivation: Currently there are no curative anticancer drugs, and drug resistance is often acquired after drug treatment. One of the reasons is that cancers are complex diseases, regulated by multiple signaling pathways and cross talks among the pathways. It is expected that drug combinations can reduce drug resistance and improve patients’ outcomes. In clinical practice, the ideal and feasible drug combinations are combinations of existing Food and Drug Administration-approved drugs or bioactive compounds that are already used on patients or have entered clinical trials and passed safety tests. These drug combinations could directly be used on patients with less concern of toxic effects. However, there is so far no effective computational approach to search effective drug combinations from the enormous number of possibilities.Results: In this study, we propose a novel systematic computational tool DrugComboRanker to prioritize synergistic drug combinations and uncover their mechanisms of action. We first build a drug functional network based on their genomic profiles, and partition the network into numerous drug network communities by using a Bayesian non-negative matrix factorization approach. As drugs within overlapping community share common mechanisms of action, we next uncover potential targets of drugs by applying a recommendation system on drug communities. We meanwhile build disease-specific signaling networks based on patients’ genomic profiles and interactome data. We then identify drug combinations by searching drugs whose targets are enriched in the complementary signaling modules of the disease signaling network. The novel method was evaluated on lung adenocarcinoma and endocrine receptor positive breast cancer, and compared with other drug combination approaches. These case studies discovered a set of effective drug combinations top ranked in our prediction list, and mapped the drug targets on the disease signaling network to highlight the mechanisms of action of the drug combinations.Availability and implementation: The program is available on request.Contact:stwong@tmhs.org

Cancer is a complex disease caused by multiple factors, which hamper effective drug discovery. Drug combination or multi target agents provide an alternate way to effectively modify disease networks. Synergistic drug pairs have special potential for treatment since they allow a desired effect to be achieved with lower total dose of administered medicine and usually with fewer side effects. However, the major challenge has been the prediction of chemotherapeutic efficacy based on the biological profile of the tumor. Because of the lack of gene expression data treated with drug combinations, in this study, we present a computational approach to identify effective drug combinations by exploiting high throughput data. We constructed tissue specific network pharmacology based on large-scale screening data on drug treatment efficacies of 130 drugs under clinical and preclinical investigation and drug-target binding affinities. We evaluated CNS cancer subset and applied logic based network algorithm to predict effective drug combinations based on drug-target interactions and single drug sensitivity profiles. Cancer cell based target inhibition network analysis in two case studies using glioma cell lines, (U87 and U251) identified distinct cell line survival pathways (p &amp;lt; 0.001), including cell proliferation, adhesion and growth factor signaling. We estimated pairwise drug synergy scores for all the target genes and identified several synergistic pairs with potential clinical relevance. Target inhibition modeling allowed systematic exploration of functional interactions between drugs and their targets to maximally inhibit multiple survival pathways. Citation Format: Uma Shankavaram, Kevin Camphausen. Target inhibitory networks and drug response modeling. [abstract]. In: Proceedings of the AACR Precision Medicine Series: Integrating Clinical Genomics and Cancer Therapy; Jun 13-16, 2015; Salt Lake City, UT. Philadelphia (PA): AACR; Clin Cancer Res 2016;22(1_Suppl):Abstract nr 42.

Complex diseases like cancer are regulated by large, interconnected networks with many pathways affecting cell proliferation, invasion, and drug resistance. However, current cancer therapy predominantly relies on the reductionist approach of one gene-one disease. Combinations of drugs may overcome drug resistance by limiting mutations and induction of escape pathways, but given the enormous number of possible drug combinations, strategies to reduce the search space and prioritize experiments are needed. In this review, we focus on the use of computational modeling, bioinformatics and high-throughput experimental methods for discovery of drug combinations. We highlight cutting-edge systems approaches, including large-scale modeling of cell signaling networks, network motif analysis, statistical association-based models, identifying correlations in gene signatures, functional genomics, and high-throughput combination screens. We also present a list of publicly available data and resources to aid in discovery of drug combinations. Integration of these systems approaches will enable faster discovery and translation of clinically relevant drug combinations.Graphical abstractSpectrum of Systems Biology Approaches for Drug Combinations.

The battle against many cancers and infectious diseases has long been hindered due to the complexity of finding potent and effective drug combinations. With each new drug considered, the number of combinations to potentially test increases exponentially, posing substantial challenges in screening throughput. These challenges further intensify when accounting for the number of doses of each drug that need to be tested in a drug combination matrix (known as matrix density). There is a pressing need to screen large amounts of combinations at sufficient density to discover new therapies for diseases like cancer, but this has traditionally been out of reach. However, the recent widespread adoption of acoustic liquid handling robots has shown promise to overcome these obstacles by allowing for intricate drug screening template designs which were previously not possible to make. Despite these advances, the throughput achieved by these technologies has been limited due to lack of broadly accessible protocols and analytical tools for drug combination screening. We present Combocat, an end-to-end platform that allows for substantial increases in throughput of drug combination screens by combining experimental protocols that can be deployed for acoustic liquid handlers, machine learning algorithms for data imputation, and software that allows for in-depth analysis of results. We first generated a reference dataset of over 250,000 unique drug combination measurements in multiple cancer cell lines. The combination data were collected in a dense format (10×10 combination matrices) using a novel drug-drug template and achieved a dramatic increase in throughput compared to conventional methods. We then used this dataset to build a computational model which allowed us to accurately estimate drug combination effects using sparse measurements and imputing non-measured values with machine learning. The sparse measurements are collected in 1536-well microplates and substantially boost the throughput capabilities of drug-drug screens. As proof-of-concept, we used our method to screen a preclinical model of neuroblastoma with 9,045 drug combinations. This represents 10% the scale of the largest drug combination studies ever reported, achieved using a fraction of the resources, and in dense formats. We validated our findings by re-screening top hits using the fully-measured, non-imputed method and demonstrate the accuracy of our platform. The Combocat platform’s documentation and codebase is open-source, and we also make a GUI available for interactive exploration of screening results. By integrating advanced experimental and computational methods, we provide a generalizable pipeline that will expedite synergy screens and the drug combination discovery process for many diseases. Citation Format: William C. Wright, Paul Geeleher. An ultrahigh-throughput synergy screening platform enables discovery of novel drug combinations [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4936.

New strategies in melanoma therapy: can nanoparticles overcome chemoresistance?

BackgroundPrioritizing candidate drugs based on genome-wide expression data is an emerging approach in systems pharmacology due to its holistic perspective for preclinical drug evaluation. In the current study, a network-based approach was proposed and applied to prioritize plant polyphenols and identify potential drug combinations in breast cancer. We focused on MEK5/ERK5 signalling pathway genes, a recently identified potential drug target in cancer with roles spanning major carcinogenesis processes.ResultsBy constructing and identifying perturbed protein–protein interaction networks for luminal A breast cancer, plant polyphenols and drugs from transcriptome data, we first demonstrated their systemic effects on the MEK5/ERK5 signalling pathway. Subsequently, we applied a pathway-specific network pharmacology pipeline to prioritize plant polyphenols and potential drug combinations for use in breast cancer. Our analysis prioritized genistein among plant polyphenols. Drug combination simulations predicted several FDA-approved drugs in breast cancer with well-established pharmacology as candidates for target network synergistic combination with genistein. This study also highlights the concept of target network enhancer drugs, with drugs previously not well characterised in breast cancer being prioritized for use in the MEK5/ERK5 pathway in breast cancer.ConclusionThis study proposes a computational framework for drug prioritization and combination with the MEK5/ERK5 signaling pathway in breast cancer. The method is flexible and provides the scientific community with a robust method that can be applied to other complex diseases.

Objectives: Small-for-gestational-age fetuses (SGA) are at high risk of intrapartum fetal compromise requiring operative delivery. In a recent study, we developed a model using a combination of three antenatal (gestational age at delivery, parity, cerebroplacental ratio) and three intrapartum (epidural use, labor induction and augmentation using oxytocin) variables for the prediction of operative delivery due to presumed fetal compromise in SGA fetuses – the Individual RIsk aSsessment (IRIS) prediction model. The aim of this study was to test the predictive accuracy of the IRIS prediction model in an external cohort of singleton pregnancies complicated by SGA.Methods: This was an external validation study using a cohort of pregnancies from two tertiary referral centers in Spain and England. The inclusion criteria were singleton pregnancies diagnosed with an SGA fetus, defined as estimated fetal weight (EFW) below the 10th centile for gestational age at 36 weeks or beyond, which had fetal Doppler assessment and available data on their intrapartum care and pregnancy outcomes. The main outcome in this study was the operative delivery for presumed fetal compromise. External validation was performed using the coefficients obtained in the original development cohort. The predictive accuracies of models were investigated with receiver operating characteristics (ROC) curves. The Hosmer–Lemeshow test was used to test the goodness-of-fit of models and calibration plots were also obtained for visual assessment. A mobile application using the combined model algorithm was developed to facilitate clinical use.Results: Four hundred twelve singleton pregnancies with an antenatal diagnosis of SGA were included in the study. The operative delivery rate was 22.8% (n = 94). The group which required operative delivery for presumed fetal compromise had significantly fewer multiparous women (19.1 versus 47.8%, p < .001 in the total study population; 19.0 versus 43.5 and 19.2 versus 49.6%, UK and Spain cohort, respectively), lower cerebroplacental ratio (CPR) multiples of median (MoM) (median: 0.77 versus 0.92, p < .001 in the total study population; 0.77 versus 0.92 and 0.77 versus 0.92, UK and Spain cohort, respectively), more inductions of labor (74.5 versus 60.1%, p = .010 in the total study population; 85.7 versus 77.2 and 71.2% and 53.1, UK and Spain cohort, respectively) and more use of oxytocin augmentation (57.4 versus 39.3%, p = .002 in the total study population; 19.0 versus 12.0 and 68.5 and 50.4%, UK and Spain cohort, respectively) compared to those who did not require operative delivery due to presumed fetal compromise. When the original antenatal model was applied to the present cohort, we observed moderate predictive accuracy (AUC: 0.70, 95% CI: 0.64–0.76), and no signs of poor fit (p = .464). The original combined model, when applied to the external cohort, had moderate predictive accuracy (AUC: 0.72, 95% CI: 0.67–0.77) and also no signs of poor fit (p = .268) without the need for refitting. A statistically significant increase in the predictive accuracy was not achieved via refitting of the combined model (AUC 0.76 versus 0.72, p = .060).Conclusions: Using our recently published model, the predictive accuracy for fetal compromise requiring operative delivery in term fetuses thought to be SGA was modest and showed no signs of poor fit in an external cohort. The IRIS tool for mobile devices has been developed to facilitate wide clinical use of this prediction model.Brief rationaleObjective: To determine the external validity of an intrapartum risk prediction model for suspected small-for-gestational age fetuses.What is already known: Small-for-gestational age fetuses are at increased risk of intrapartum compromise. Fetal weight alone is a poor marker for adverse outcomes and a comprehensive prediction model has been previously suggested.What this study adds: Multivariable prediction model showed good accuracy and calibration in this external validation study. The significance of some variables was different between the original and external validation cohort and there was a small margin for improvement with model refitting. A mobile application has been developed to facilitate clinical use.

The early determination of response to neoadjuvant therapy (NAT) in triple-negative breast cancer would enable the treating oncologist to adapt the therapeutic regimen of a non-responding patient (e.g., by changing dosage, dose schedule, prescribed drugs), and thereby improve treatment outcomes while avoiding unnecessary toxicities. To address this challenge, we propose to use personalized, in silico forecasts of tumor response to therapeutic regimens via a mechanistic mathematical model calibrated with patient-specific longitudinal multi-parametric magnetic resonance imaging (MRI) data acquired early in the course of NAT. Here, we extend our mechanistic model to include a new term describing the synergistic effects of NAT drug combinations and identify the driving parameters involved in its formulation by means of a sensitivity analysis. Our model describes tumor cell dynamics as a combination of proliferation, which is regulated by a logistic term, and mobility, which is described as a diffusion process constrained by the local tumor-induced mechanical stress. Tumor cell density is extracted from diffusion-weighted MRI data, while tissue mechanical properties are defined from segmented T1-weighted MRI data. We adjust the tumor proliferation rate in response to NAT drug combinations with a recent model of drug synergy, MuSyC, which accounts for distinct types of synergistic drug effects (synergy of potency vs. synergy of efficacy). We also consider the heterogeneous intratumoral delivery of drugs by means of perfusion maps estimated from dynamic contrast-enhanced MRI data. We use Sobol’s method for the sensitivity analysis of two different tumors - one well-perfused and one poorly-perfused. We simulate a four-cycle NAT protocol in which NAT drugs are delivered every 14 days, and assess the total effect (ST) of each parameter on the mean relative difference of tumor cell density with respect to a control simulation of tumor growth without NAT. Sensitivity analysis results directly depend on the definition of the parameter space, which we construct by combining two approaches. First, we experimentally constrain parameter ranges using time-resolved, high-throughput, automated microscopy assays to capture the changes in proliferation rates of various breast cancer lines (HCC1143, SUM149, MDAMB231, and MDAMB468) caused by two standard drug combinations: paclitaxel with carboplatin and doxorubicin with perfosfamide (metabolic derivative of the pro-drug cyclophosphamide), and fitting the MuSyC model to these data. Second, we scale the resulting in vitro parameter ranges to clinically-relevant in vivo ranges by running an in silico study with our mechanistic model of breast cancer growth and NAT response. Our results show that, out of the ten parameters involved in the synergy term, three have a dominant role in the dynamics of breast cancer during NAT (ST &amp;gt; 0.1): synergistic potency, the maximal change in tumor cell proliferation by the slowest decaying drug, and its concentration producing half of maximal effects. The other parameters have marginal (0.02 &amp;lt; ST &amp;lt; 0.1) to negligible effect (ST &amp;lt; 0.02). Ongoing studies are assessing the ability of our mechanistic model to forecast NAT response over a small patient cohort after patient-specific calibration of the driving parameters identified in the present study. Citation Format: Guillermo Lorenzo, Angela M. Jarrett, Christian T. Meyer, Darren R. Tyson, Vito Quaranta, Thomas E. Yankeelov. Identifying relevant parameters that characterize the early response to NAT in breast cancer patients using a novel personalized mechanistic model integrating in vitro and in vivo imaging data [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PS13-44.

Sequential courses of anticancer target therapy lead to selection of drug-resistant cells, which results in continuous decrease of clinical response. Here we present a new approach for predicting effective combinations of target drugs, which act in a synergistic manner. Synergistic combinations of drugs may prevent or postpone acquired resistance, thus increasing treatment efficiency. We cultured human ovarian carcinoma SKOV-3 and neuroblastoma NGP-127 cancer cell lines in the presence of Tyrosine Kinase Inhibitors (Pazopanib, Sorafenib, and Sunitinib) and Rapalogues (Temsirolimus and Everolimus) for four months and obtained cell lines demonstrating increased drug resistance. We investigated gene expression profiles of intact and resistant cells by microarrays and analyzed alterations in 378 cancer-related signaling pathways using the bioinformatical platform Oncobox. This revealed numerous pathways linked with development of drug resistant phenotypes. Our approach is based on targeting proteins involved in as many as possible signaling pathways upregulated in resistant cells. We tested 13 combinations of drugs and/or selective inhibitors predicted by Oncobox and 10 random combinations. Synergy scores for Oncobox predictions were significantly higher than for randomly selected drug combinations. Thus, the proposed approach significantly outperforms random selection of drugs and can be adopted to enhance discovery of new synergistic combinations of anticancer target drugs.

Estimation of predictive accuracy in survival analysis using R and S-PLUS