ACLPred: an explainable machine learning and tree-based ensemble model for anticancer ligand prediction.

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Growing cancer incidence necessitates novel therapeutic agents.
• Traditional experimental drug screening is resource-intensive.
• Machine learning offers a rapid, cost-effective alternative for identifying anticancer compounds.

Purpose of the Study:

• To develop and validate a machine learning model for predicting anticancer activity of small molecules.
• To identify key molecular features contributing to anticancer properties.
• To create an accessible tool for researchers to screen potential drug candidates.

Main Methods:

• Training classification models using molecular descriptors of known anticancer and non-anticancer compounds.
• Applying multistep feature selection to identify significant molecular descriptors.
• Employing and evaluating various machine learning algorithms, including Light Gradient Boosting Machine (LGBM).
• Utilizing SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• The LGBM model achieved a prediction accuracy of 90.33% and an AUROC of 97.31%.
• The developed tool, ACLPred, demonstrated superior prediction accuracy and generalizability over existing methods.
• SHAP analysis indicated that topological molecular features significantly influenced the model's predictions.

Conclusions:

• Machine learning, particularly the LGBM algorithm implemented in ACLPred, provides an effective and accurate method for identifying potential anticancer compounds.
• ACLPred offers a user-friendly, open-source solution for accelerating anticancer drug discovery.
• Topological features are crucial for predicting anticancer activity, offering insights for future drug design.