MLASM: Machine learning based prediction of anticancer small molecules.

Area of Science:

• Computational chemistry and bioinformatics
• Drug discovery and development

Background:

• Cancer is a leading global cause of death, necessitating effective treatments.
• Anticancer peptides (ACPs) show therapeutic promise but experimental identification is costly and time-consuming.
• Anticancer small molecules (ACSMs) offer an alternative therapeutic strategy.

Purpose of the Study:

• To develop a machine learning (ML) model for predicting anticancer small molecules (ACSMs).
• To overcome the limitations of experimental identification of potential anticancer agents.
• To evaluate the performance of various ML algorithms in predicting ACSMs.

Main Methods:

• Developed ML models using five algorithms: Random Forest (RF), LightGBM, K-nearest neighbors (KNN), Decision Tree (DT), and Extreme Gradient Boosting (XGB).
• Trained models on a dataset of 10,000 compounds.
• Validated model performance using a test set and external validation with FDA-approved anticancer drugs.

Main Results:

• Identified RF, LightGBM, and XGB as the top three performing models.
• The LightGBM model achieved the highest accuracy (79%) and Area Under the Curve (AUC) of 0.88.
• LightGBM correctly predicted 9 out of 10 active compounds in external validation, outperforming RF (8/10) and XGB (7/10).

Conclusions:

• The developed ML models, particularly LightGBM, show robust prediction capabilities for identifying anticancer small molecules.
• This ML-based approach offers a more efficient and cost-effective alternative to experimental methods for drug discovery.
• Machine learning holds significant potential for advancing cancer treatment research and accelerating the development of novel anticancer therapies.