Development of machine learning classifiers to predict compound activity on prostate cancer cell lines

Area of Science:

• Oncology
• Computational Chemistry
• Bioinformatics

Background:

• Prostate cancer is a leading cancer in men, with effective treatments available for early stages.
• Aggressive prostate cancer variants lack sufficient therapeutic options, necessitating novel drug discovery.
• Predictive models can accelerate the identification of potential antiproliferative agents.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the activity of compounds against prostate cancer cell lines (PC-3 and DU-145).
• To identify optimal machine learning algorithms and data thresholds for accurate prediction of compound activity.
• To explore potential biological targets for future drug design based on compound activity annotations.

Main Methods:

• Utilized homogeneous cell-based antiproliferative assay data for PC-3 and DU-145 prostate cancer cell lines.
• Developed and compared 10 different machine learning algorithms, fine-tuning data thresholds and feature selection.
• Employed metrics like Matthews Correlation Coefficient (MCC) to evaluate model performance.

Main Results:

• Achieved prediction performances with MCC values exceeding 0.60 for both PC-3 and DU-145 cell lines.
• Developed in silico models combining data from both cell lines, demonstrating excellent precision.
• Identified associations between ligand activity and potential biological targets.

Conclusions:

• Machine learning models can effectively predict antiproliferative activity against prostate cancer cell lines.
• Optimized models provide a valuable tool for screening potential therapeutics for aggressive prostate cancer.
• Further investigation into identified biological targets may lead to the development of more effective prostate cancer drugs.