Drug Sensitivity Prediction Using Machine Learning on Integrated COSMIC, DGIdb, and GDSC Data

Aug 3, 2026

Uploaded by Ophelia

Drug sensitivity prediction links drug efficacy to mutational profiles to support effective treatment of complex diseases like cancer, where ongoing mutations can alter how tumors respond to therapies. The study performs statistical analysis of drug–disease–gene interactions and builds a general processing pipeline with machine learning models to predict cancer-cell drug sensitivity from genetic mutations. Four open-source datasets (drug sensitivity, somatic mutations, and gene–drug interactions) are integrated into an enriched database with preprocessing for text encoding, filtering, and optimization, followed by interaction testing, contribution quantification, ablation, and feature-importance analyses; results show strong generalizability with high R2 values across independent sources.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Abstract

Introduction

Drug repurposing and therapy response variation

Drug resistance and sensitivity concepts

Data sources: GDSC, COSMIC, DGIdb, and related resources

Motivation for computational drug response prediction

Received 12 January 2026, accepted 25 January 2026, date of publication 29 January 2026, date of current version 5 February 2026. Digital Object Identifier 10.1109/ACCESS.2026.3659340 Drug Sensitivity Prediction Using Machine Learning on Integrated COSMIC, DGIdb, and GDSC Data BERFU MERGEN 1, MERAL COBAN2, SENAY SUEDA OZKAN2, OMER FARUK BASARAN2, AND GIYASETTIN OZCAN1,2 1Department of Translational Medicine, Bursa Uludağ University, 16059 Bursa, Türkiye 2Department of Computer Engineering, Bursa Uludağ University, 16059 Bursa, Türkiye Corresponding author: Omer Faruk Basaran ([omerbasaran@uludag.edu.tr](omerbasaran@uludag.edu.tr)) The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. ABSTRACT Analyzing the relationship between drug efficacy and sensitivity to mutational profiles is necessary for the effective treatment of complex diseases such as cancer. Particularly, cancerous tissues undergo constant change as a result of ongoing mutations, and the sensitivity of drugs to cancer may change as a result of new mutations. For this purpose, this study aims to present a statistical analysis of drug–disease– gene interactions. Furthermore, a general processing pipeline and machine learning models were developed to predict the drug sensitivity of cancer cells according to genetic mutations. To achieve this, four well-known open-source databases, including drug sensitivity data from cancer cell lines, two somatic mutation data resources, and a gene-drug interaction database, were integrated to assess an enriched database. Next, various preprocessing techniques, including text encoding, filtering, and optimization, were implemented to attain an efficient new dataset for statistical analysis and machine learning. Statistical analyses were conducted to investigate gene–drug interactions on the enriched database and to quantify their relative contributions to drug sensitivity. On the other hand, developed machine learning models predict drug sensitivity from somatic mutation or drug interaction datasets. The research also includes ablation studies and feature importance to introduce a thorough analysis of gene and drug sensitivity. The developed pipeline not only yielded an R2 of 0 .91 in initial evaluations but also demonstrated robust generalizability by maintaining a 0 .73 R2 score in predicting AUC values across independent data sources. Overall, statistical analysis, machine learning performances, and ablation studies offer a new perspective on drug sensitivity prediction. INDEX TERMS Drug sensitivity, prediction of Z-score by machine learning, integration of multiple data resources, data manipulation, machine learning models. I. INTRODUCTION Rapid advancements in molecular biology, genomics, and computational technologies over the past thirty years have laid the foundation for major progress in pharmacological research. The integration of these fields has particularly accelerated developments in drug discovery and drug repurposing. [1],[2],[3] . In terms of drug studies, recent research focused on drug discovery and drug repurposing [4] . Drug repurposing enables known drugs to gain new therapeutic uses beyond The associate editor coordinating the review of this manuscript and approving it for publication was Olarik Surinta . their defined purposes [4], [5] . Furthermore, drug repurposing aims to reduce the costs, time, and effort associated with drug development by bypassing early-stage processes such as discovery, preclinical testing, and production preparation [6] . For example, drugs such as Minoxidil, Thalidomide, and Sildenafil, which were previously used to treat different conditions, have been repurposed and are now used to treat different diseases [4],[7],[8] . When we delve deeper into the topic of drug repurposing, we inevitably encounter the concepts of drug resistance and drug sensitivity. Drug resistance refers to the inability

FAQs

What is the main goal of the study?

To predict cancer-cell drug sensitivity from genetic mutations by integrating multiple public datasets and combining statistical analysis with machine learning models.

Which data sources are integrated to build the enriched dataset?

Four open-source resources are integrated, including drug sensitivity data from cancer cell lines, two somatic mutation datasets, and a gene–drug interaction database.

How do the authors evaluate the prediction performance?

They conduct statistical analyses and compute predictive effectiveness, reporting strong initial performance (R2) and maintaining robust generalizability by achieving an R2 around 0.73 for AUC prediction across independent data sources, supported by ablation and feature-importance analyses.

Aug 3, 2026

Uploaded by Ophelia