Support Vector Machine-Based Prediction Models for Drug Repurposing and Designing Novel Drugs for Colorectal Cancer

Area of Science:

• Computational chemistry
• Pharmacology
• Oncology

Background:

• Colorectal cancer (CRC) incidence is rising, presenting a significant therapeutic challenge.
• There is a critical need for novel drug therapies to improve CRC treatment outcomes.

Purpose of the Study:

• To develop robust quantitative structure-activity relationship (QSAR) models for predicting anticancer compounds for colorectal cancer (CRC).
• To identify novel and existing drugs with potential efficacy against CRC using computational methods.

Main Methods:

• Utilized pharmacological profiles of anticancer drugs from the Genomics of Drug Sensitivity in Cancer (GDSC) database.
• Developed QSAR models using the Support Vector Machine (SVM) algorithm.
• Validated models using 10-fold cross-validation and analyzed chemical descriptors and drug-to-oncogene relationships.

Main Results:

• Achieved high QSAR model performance (R² 0.609–0.827) across 12 CRC cell lines, with peak performance for SW1417 (R²=0.827) and GP5d (R²=0.786).
• Identified KRFP314 and KRFPC314 as prevalent chemical descriptors correlating with drug activity.
• Identified viomycin and diamorphine as potential anti-CRC drugs.

Conclusions:

• Developed validated QSAR models for predicting anti-CRC drug efficacy.
• Integrated models into an accessible web server, "ColoRecPred", for screening potential CRC drug candidates.
• The study provides a valuable tool for accelerating the discovery of new colorectal cancer treatments.