MMCL-CDR: enhancing cancer drug response prediction with multi-omics and morphology images contrastive representation learning

BackgroundPredicting patient drug response based on a patient’s molecular profile is one of the key goals of precision medicine in breast cancer (BC). Multiple drug response prediction models have been developed to address this problem. However, most of them were developed to make sensitivity predictions for multiple single drugs within cell lines from various cancer types instead of a single cancer type, do not take into account drug properties, and have not been validated in cancer patient-derived data. Among the multi-omics data, gene expression profiles have been shown to be the most informative data for drug response prediction. However, these models were often developed with individual genes. Therefore, this study aimed to develop a drug response prediction model for BC using multiple data types from both cell lines and drugs.MethodsWe first collected the baseline gene expression profiles of 49 BC cell lines along with IC50 values for 220 drugs tested in these cell lines from Genomics of Drug Sensitivity in Cancer (GDSC). Using these data, we developed a multiple-layer cell line-drug response network (ML-CDN2) by integrating a one-layer cell line similarity network based on the pathway activity profiles and a three-layer drug similarity network based on the drug structures, targets, and pan-cancer IC50 profiles. We further used ML-CDN2 to predict the drug response for new BC cell lines or patient-derived samples.ResultsML-CDN2 demonstrated a good predictive performance, with the Pearson correlation coefficient between the observed and predicted IC50 values for all GDSC cell line-drug pairs of 0.873. Also, ML-CDN2 showed a good performance when used to predict drug response in new BC cell lines from the Cancer Cell Line Encyclopedia (CCLE), with a Pearson correlation coefficient of 0.718. Moreover, we found that the cell line-derived ML-CDN2 model could be applied to predict drug response in the BC patient-derived samples from The Cancer Genome Atlas (TCGA).ConclusionsThe ML-CDN2 model was built to predict BC drug response using comprehensive information from both cell lines and drugs. Compared with existing methods, it has the potential to predict the drug response for BC patient-derived samples.

The selection of optimal therapies for patients with ovarian cancer is difficult, especially in recurrent disease. An implantable microdevice (IMD) is a new technology to evaluate in situ drug responses in cancer. An IMD, about the size of a grain of rice, is implanted into human or murine solid tumors and releases microdoses of up to 20 compounds in spatially distinct regions of the tumor. Histologic analysis of tissue surrounding the IMD is used to quantitate tumor response. Using IMDs, we can measure tumor sensitivity to multiple drugs or drug combinations in parallel, directly in the tumor with its native microenvironment. Previous studies showed that measurement of local drug response with IMDs in solid tumors correlates with systemic drug response in animal models. We are conducting a clinical trial (NCT04701645) to evaluate the safety and feasibility of IMD implantation in patients with ovarian cancer undergoing surgical resection for clinical indications. Several IMDs, each with 16 standard-of-care drugs and controls, are implanted into a tumor 24-72 hours prior to surgery using a percutaneous delivery needle under CT guidance. After the mass is resected, the tumor tissue containing the IMDs is formalin-fixed, sectioned, and stained to evaluate cellular responses to drug treatment. The aim of the trial is to assess the safety and the feasibility of IMDs in ovarian cancer. Four patients have enrolled to date (histologies: 2 high-grade serous, 1 clear cell, 1 granulosa cell) with 3-4 IMDs implanted per patient. No adverse events attributable to the IND were observed. Of 14 total devices, 10 provided evaluable histology information (4 were dislodged from the tumor ex vivo during pathology processing). Several assays are being applied to evaluate the effects of each drug on the tumor tissue, including 1) immunohistochemistry for apoptosis (cleaved caspase 3, CC3); 2) cyclic immunofluorescence for protein markers; 3) spatial transcriptomics. Regions of interest are selected adjacent to each drug reservoir, and imaging methods quantitate different cell types and markers of drug response. Initial results from three patients demonstrate moderate induction of apoptosis via CC3 staining with a subset of drugs, but minimal CC3 staining for other drugs, in each patient’s tumor. Cyclic immunofluorescence staining is completed on tissue from the first two patients, and a spatial transcriptomics panel is completed on tissue from one patient, both technically successful, and data analysis is ongoing. Additional patients are being enrolled to the clinical trial. In summary, our clinical trial experience indicates that IMDs are safe to implant into tumors of patients with ovarian cancer and that it is feasible to retrieve the IMDs after surgery and obtain informative measurements of mRNA and protein responses in ovarian cancers exposed to different drugs. IMDs may represent an effective new technology for functional precision medicine in ovarian cancer. Citation Format: Elizabeth H. Stover, Sharath K. Bhagavatula, Christine A. Dominas, Sebastian W. Ahn, Samantha E. Martin, David L. Kolin, Michelle S. Hirsch, Madeline Polak, Alexis Rabbitt, Jeanette Gardner, Kimberley MacNeill, Zuzana Tatarova, Michael Worley Jr., Kevin M. Elias, Joyce F. Liu, Ursula A. Matulonis, Stuart G. Silverman, Neil S. Horowitz, Oliver Jonas. Clinical trial of an implantable microdevice to evaluate drug responses in ovarian cancer [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Functional and Genomic Precision Medicine in Cancer: Different Perspectives, Common Goals; 2025 Mar 11-13; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2025;85(5 Suppl):Abstract nr B012.

DRN-CDR: A cancer drug response prediction model using multi-omics and drug features

SummaryPhenotypic heterogeneity in cancer is often caused by different patterns of genetic alterations. Understanding such phenotype-genotype relationships is fundamental for the advance of personalized medicine. We develop a computational method, named NETPHIX (NETwork-to-PHenotype association with eXclusivity) to identify subnetworks of genes whose genetic alterations are associated with drug response or other continuous cancer phenotypes. Leveraging interaction information among genes and properties of cancer mutations such as mutual exclusivity, we formulate the problem as an integer linear program and solve it optimally to obtain a subnetwork of associated genes. Applied to a large-scale drug screening dataset, NETPHIX uncovered gene modules significantly associated with drug responses. Utilizing interaction information, NETPHIX modules are functionally coherent and can thus provide important insights into drug action. In addition, we show that modules identified by NETPHIX together with their association patterns can be leveraged to suggest drug combinations.

Prediction of drug response in multilayer networks based on fusion of multiomics data.

BackgroundDrug resistance is a critical obstacle in cancer therapy. Discovering cancer drug response is important to improve anti-cancer drug treatment and guide anti-cancer drug design. Abundant genomic and drug response resources of cancer cell lines provide unprecedented opportunities for such study. However, cancer cell lines cannot fully reflect heterogeneous tumor microenvironments. Transferring knowledge studied from in vitro cell lines to single-cell and clinical data will be a promising direction to better understand drug resistance. Most current studies include single nucleotide variants (SNV) as features and focus on improving predictive ability of cancer drug response on cell lines. However, obtaining accurate SNVs from clinical tumor samples and single-cell data is not reliable. This makes it difficult to generalize such SNV-based models to clinical tumor data or single-cell level studies in the future.ResultsWe present a new method, DualGCN, a unified Dual Graph Convolutional Network model to predict cancer drug response. DualGCN encodes both chemical structures of drugs and omics data of biological samples using graph convolutional networks. Then the two embeddings are fed into a multilayer perceptron to predict drug response. DualGCN incorporates prior knowledge on cancer-related genes and protein–protein interactions, and outperforms most state-of-the-art methods while avoiding using large-scale SNV data.ConclusionsThe proposed method outperforms most state-of-the-art methods in predicting cancer drug response without the use of large-scale SNV data. These favorable results indicate its potential to be extended to clinical and single-cell tumor samples and advancements in precision medicine.

BackgroundHuman cancer cell lines are used in research to study the biology of cancer and to test cancer treatments. Recently there are already some large panels of several hundred human cancer cell lines which are characterized with genomic and pharmacological data. The ability to predict drug responses using these pharmacogenomics data can facilitate the development of precision cancer medicines. Although several methods have been developed to address the drug response prediction, there are many challenges in obtaining accurate prediction.MethodsBased on the fact that similar cell lines and similar drugs exhibit similar drug responses, we adopted a similarity-regularized matrix factorization (SRMF) method to predict anticancer drug responses of cell lines using chemical structures of drugs and baseline gene expression levels in cell lines. Specifically, chemical structural similarity of drugs and gene expression profile similarity of cell lines were considered as regularization terms, which were incorporated to the drug response matrix factorization model.ResultsWe first demonstrated the effectiveness of SRMF using a set of simulation data and compared it with two typical similarity-based methods. Furthermore, we applied it to the Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE) datasets, and performance of SRMF exceeds three state-of-the-art methods. We also applied SRMF to estimate the missing drug response values in the GDSC dataset. Even though SRMF does not specifically model mutation information, it could correctly predict drug-cancer gene associations that are consistent with existing data, and identify novel drug-cancer gene associations that are not found in existing data as well. SRMF can also aid in drug repositioning. The newly predicted drug responses of GDSC dataset suggest that mTOR inhibitor rapamycin was sensitive to non-small cell lung cancer (NSCLC), and expression of AK1RC3 and HINT1 may be adjunct markers of cell line sensitivity to rapamycin.ConclusionsOur analysis showed that the proposed data integration method is able to improve the accuracy of prediction of anticancer drug responses in cell lines, and can identify consistent and novel drug-cancer gene associations compared to existing data as well as aid in drug repositioning.

Introduction: The integration of cell line features and drug features in computational Cancer Drug Response (CDR) prediction methods enables a nuanced understanding of cellular responses and drug effects, which may lead to improvements in drug discovery and precision oncology. It helps identify promising drug candidates for experimental validation, avoid treatments that are unlikely to benefit a patient, and reduce unnecessary exposure to toxic drugs. Methods: In this paper, we propose DCGPert-CDR, which integrates drug structural features, cell line multi-omics data, and target gene perturbation profiles for predicting IC50 responses. The methodology involves the extraction of cell line multi-omics data, including genomics, transcriptomics, and epigenomics, together with the molecular structural features of the drug. The gene perturbation profiles are computed from transcriptional changes of the prioritized target genes before and after the drug treatment. A graph clustering approach, followed by network propagation, is applied to prioritize drug target genes. The resultant feature vectors are concatenated and fed into a prediction module, consisting of a ResNet, which predicts the IC50 values of drugs across various cancer cell lines. Results: DCGPert-CDR produces promising results when compared to similar methods, with Pearson’s correlation rp of 0.841 and Spearman’s correlation rs of 0.786 computed between predicted and actual IC50 values, while for other methods, rp was in the range of 0.768 to 0.8183 and rs was between 0.735 to 0.757. Drugs such as Foretinib, Crizotinib, Tivozanib, SNX-2112, and PHA- 665752 are found to be most sensitive after analyzing the predicted response values across various cancer cell lines. Discussion: The improved results indicate that the proposed method effectively predicts responses that closely match the known IC50 values. Case studies are conducted on 24 TCGA cancer types, also revealing sensitive drugs for each cancer type, which are corroborated with clinical evidence. Dependence on the availability of drug and cell line data, as well as the absence of real-time data validation, remains a key challenge. Conclusion: The method can reliably capture the relationship between drugs and cell lines, indicating its potential utility in predicting drug sensitivity. The method effectively identified the most sensitive drugs among individual cancer types.

The APOBEC (apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like) gene family of cytidine deaminases includes evolutionarily conserved genes that play important roles in DNA repair and mRNA editing. Activity of at least two APOBEC family members, APOBEC3A and APOBEC3B, can lead to kataegis, a mutagenic process in cancer cells that generates clusters of closely spaced, single strand specific C->T DNA substitutions. APOBEC mutagenesis has a characteristic signature, most commonly represented by the 5’-Tp(C->T)pW-3’ sequence motif, with additional substitutions also reported. This hypermutation signature and high mRNA expression of APOBEC3A and APOBEC3B have been associated with several cancer types. Most previous studies of APOBEC signatures have examined tumor sequence data from clinical samples, for which limited or no information about drug response was available. We investigated the presence of the mutational signature and mRNA expression patterns of the APOBEC3A and APOBEC3B genes in extensively characterized cell lines, in order to identify those cell lines that carry mutations generated by kataegis, with the aim of establishing associations between the APOBEC mutational signature, individual cancer types, and the patterns of sensitivity to antitumor agents. For this purpose, we analyzed whole exome sequencing (WES) data and mRNA expression of the APOBEC3A and APOBEC3B genes in two resources with extensive drug response data: the NCI-60 cell line panel, which includes 59 human cancer cell lines representing 9 cancer types and drug response information for thousands of anticancer agents, and the Cancer Cell Line Encyclopedia (CCLE), which provides WES, whole genome, and RNA-seq sequence information on hundreds of cancer cell lines and drug response data to over 200 agents. We analyzed WES data of 325 CCLE cell lines and 59 NCI-60 cell lines, with variants identified using GATK pipeline and Varscan2 software. The variants in each cell line were filtered to remove common polymorphisms in dbSNP and 1000 Genome Project databases. We searched the discovered variants for the presence of APOBEC signatures, 5’-Tp(C->AGT)pN-3’, 5’-Tp(C->AGT)pD-3’, and 5’-Tp(C->AGT)pW-3’ in closely spaced (1000 and 10,000 bp) windows that appeared on the same DNA strand. We will discuss the use of optimal filters for detecting APOBEC mutational signatures and will present the analyses of associations between APOBEC signatures, mutational load of the tumor cell lines, APOBEC gene expression, and chemosensitivity to treatment. These results contribute to additional characterization of available cell lines by providing information about specific mutational signatures in different categories of cancer. Our findings may assist with identifying antitumor agents that would be appropriate for treatment of cancer cells with specific signature patterns generated by APOBEC mutagenesis. Citation Format: Suleyman Vural, Julia Krushkal, Richard Simon. Analysis of APOBEC3A and APOBEC3B mutational signatures using next-generation sequencing data from cancer cell lines [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 2590. doi:10.1158/1538-7445.AM2017-2590

Differential Splicing of Skipped Exons Predicts Drug Response in Cancer Cell Lines

Prediction of cancer drug response based on heterogeneous graph neural networks and multi-omics data.

Robust prediction of in vivo chemotherapeutic response, using baseline gene expression and drug sensitivity data gathered on cancer cell lines, has been a profoundly important, long standing and controversial problem in pharmacogenomics. Here, we present for the first time, a solution to this problem. Currently, personalizing cancer chemotherapy relies on pathology and more recently molecular biomarker-based approaches (e.g. ERBB2 amplification in breast cancer). However, as the driving biology are normally not fully understood, the majority of existing biomarkers do not capture a substantial proportion of variability in drug response. This partly explains the commonly observed lack of reproducibility of findings (e.g. from many conventional gene expression signatures) when these markers are applied to new datasets. In this study, we developed an approach to predict in vivo drug sensitivity that leverages whole-genome gene expression microarray data and allows the expression of every gene to influence the prediction by a small amount. The method works by fitting a ridge regression model of baseline genome-wide gene expression levels against in vitro drug sensitivity in a very large panel of approximately 700 cancer cell lines. Then, after a (crucial) data homogenization step, these models are applied to baseline expression levels from primary tumor biopsies. Our method successfully predicted patient response to different chemotherapeutic agents in three (of four total suitable) independent, publicly available clinical trials, each investigating different drugs and different types of cancer. In each of these cases, we predicted drug response at least as accurately as previously published models that had been derived from the clinical data itself. Interestingly, our approach could also predict clinical response in the absence of any known drug sensitivity biomarker. We effectively enriched for drug responders in breast, myeloma and lung cancers, treated with docetaxel, bortezomib and erlotinib respectively, thus identifying responders to both cytotoxic and targeted agents. Many previous clinical trials and in vitro assays have attempted to discover biomarkers of drug sensitivity, but found that the genes/aberrations which they had identified, performed poorly as predictors, once applied to out-of-batch sets of samples. Our models, on the other hand, are trained on an independent set of cancer cell lines and performed well on three completely separate and independent clinical trial datasets (all assessed using different microarray platforms). These results have far-reaching implications for personalized medicine and drug development (e.g. for the development of companion diagnostics). All datasets and bioinformatics tools to reproduce our results are publicly available. Citation Format: Paul Geeleher, Nancy Cox, R. Stephanie Huang. Clinical drug response can be predicted using baseline gene expression levels and in vitro drug sensitivity in cell lines. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 5561. doi:10.1158/1538-7445.AM2014-5561

The development of computational methods for the prediction of effective therapeutic strategies based on the genomic information of patients is the main challenge of precision medicine. Since the 21st century, next-generation sequencing (NGS) has opened up new possibilities for personalized medicine. Extensive characterization at the molecular level for hundreds of cancer cell lines has been brought to the public eye by many organizations and agencies around the world. For example, the National Cancer Institute 60 Human Cancer Cell Line Screen (NCI-60), Cancer Cell Line Encyclopedia (CCLE) and Genomics of Drug Sensitivity in Cancer (GDSC) have provided large-scale omics data such as genomic, transcriptomic and epigenomic data characterizing cancer cell lines, and The Cancer Genome Atlas (TCGA) has molecularly characterized over 20000 primary cancers of patients. Combined with the drug response data of cancer cell lines, multiomics data could be used to analyse the mechanisms of action of anticancer drugs, which could be incorporated into precision medicine strategies. Over several decades, artificial intelligence (AI) technologies based on big data have revolutionized bioinformatics. AI has built a bridge between genomics and drug sensitivity by promoting the development of predictive models for the drug response of cancer cell lines. The 2012 NCI-DREAM drug prediction challenge has been particularly influential, as the innovative applications of machine learning that emerged from it have laid the groundwork for future studies. However, classic machine learning models are still challenging in terms of predictability because they limit the systematic integration of high-dimensional multiomics data. Therefore, network-based approaches, including link prediction and network representation, have become mainstream methods for drug response prediction. On the one hand, network-based approaches have not faced the “small n, large p” problem since the multiomics features are either represented in a gene/protein network or embedded in similarity networks between cell lines. On the other hand, the introduction of gene regulatory networks (GRNs) and protein-protein interactions (PPIs) into the predictive model can provide a functional background for the integration of genomic data and thereby improve the predictive performance of drug response. In addition to network-based approaches, multimodal deep learning models can systematically integrate multiomic data by considering them as different modalities. Generally, there are three feature fusion methods in deep neural networks: Input-level feature fusion (early fusion), intermediate feature fusion and decision-level fusion (late fusion). Intermediate feature fusion is predominant in drug response prediction studies, by which features are learned separately for each type of omics data and then integrated into one unified representation to be used as the input for a classifier or a regressor. Moreover, the features of drug structures can be used as a model to improve the performance. In brief, we summarize the characteristics of publicly accessible genomic databases and discuss the trends of artificial intelligence applications in drug sensitivity prediction for cancer cell lines, including machine learning, networks and multimodal deep neural networks.

Accurate prediction of cancer drug responses is essential for advancing precision oncology. This work aims to improve generalization and robustness in drug response prediction by modeling complex drug-cell line interactions. We propose HypergraphCDR, a Hypergraph Convolutional Network model with Hypergraph Contrastive Learning for cancer drug response prediction. Multiomics features of cancer cell lines are first compressed using an autoencoder. A Hypergraph is then constructed to capture high-order relationships between drugs and cancer cell lines, with Hypergraph Convolutional Networks generating drug embeddings. In parallel, cell line embeddings are generated via a neural network. Drug and cell line embeddings are jointly optimized and integrated into a regression framework using a combination of supervised regression loss and contrastive loss. Extensive experiments demonstrate that HypergraphCDR consistently outperforms state-of-the-art methods in terms of Pearson Correlation Coefficient ($\mathit {PCC}$), Spearman Correlation Coefficient ($\mathit {SCC}$), and Coefficient of Determination ($\mathit {R}^{2}$). Moreover, in independent experiments involving unseen drugs and unseen cell lines, as well as tissue-specific evaluations, HypergraphCDR exhibits superior generalization performance compared with recent baselines. By explicitly modeling higher-order drug-cell line relationships, HypergraphCDR substantially enhances the accuracy and robustness of cancer drug response prediction. This study provides an effective and generalizable computational framework for drug response prediction, supporting reliable drug screening and treatment strategy development in precision medicine.

Prediction of the response of cancer patients to different treatments and identification of biomarkers of drug response are two major goals of individualized medicine. Here, we developed a deep learning framework called TINDL, completely trained on preclinical cancer cell lines (CCLs), to predict the response of cancer patients to different treatments. TINDL utilizes a tissue-informed normalization to account for the tissue type and cancer type of the tumors and to reduce the statistical discrepancies between CCLs and patient tumors. Moreover, by making the deep learning black box interpretable, this model identifies a small set of genes whose expression levels are predictive of drug response in the trained model, enabling identification of biomarkers of drug response. Using data from two large databases of CCLs and cancer tumors, we showed that this model can distinguish between sensitive and resistant tumors for 10 (out of 14) drugs, outperforming various other machine learning models. In addition, our small interfering RNA (siRNA) knockdown experiments on 10 genes identified by this model for one of the drugs (tamoxifen) confirmed that tamoxifen sensitivity is substantially influenced by all of these genes in MCF7 cells, and seven of these genes in T47D cells. Furthermore, genes implicated for multiple drugs pointed to shared mechanism of action among drugs and suggested several important signaling pathways. In summary, this study provides a powerful deep learning framework for prediction of drug response and identification of biomarkers of drug response in cancer. The code can be accessed at https://github.com/ddhostallero/tindl.