SMIREP: predicting chemical activity from SMILES.

Area of Science:

• Computational Chemistry
• Cheminformatics
• Machine Learning

Background:

• Traditional structure-activity-relationship (SAR) prediction involves separate fragment generation and machine learning modeling steps.
• These conventional methods often result in highly accurate but difficult-to-interpret predictive models.

Purpose of the Study:

• To introduce SMIREP, a novel SAR algorithm that integrates fragment and model generation.
• To demonstrate SMIREP's ability to produce simple, interpretable IF-THEN rules containing readily understandable SMILES fragments.

Main Methods:

• SMIREP combines the IREP rule learner with a novel fragmentation algorithm for SMILES strings.
• The algorithm was evaluated on three distinct prediction tasks: estrogen receptor binding, mutagenicity, and biodegradability.

Main Results:

• SMIREP successfully generated easily interpretable rules across all evaluated applications.
• The predictive accuracies achieved by SMIREP were comparable to existing state-of-the-art techniques.

Conclusions:

• SMIREP offers a valuable alternative for SAR prediction by providing interpretable models without sacrificing accuracy.
• The algorithm's integrated approach simplifies the SAR prediction workflow for computational chemists.