Applied Machine Learning Toward Drug Discovery Enhancement: Leishmaniases as a Case Study

Area of Science:

• Computational chemistry and cheminformatics
• Machine learning in drug discovery
• Parasitology and infectious diseases

Background:

• Drug discovery for leishmaniasis faces high attrition rates, necessitating advanced computational approaches.
• Machine learning (ML) combined with chemoinformatics offers a powerful strategy to enhance drug discovery efficiency.
• Identifying novel anti-leishmanial agents is crucial for combating leishmaniasis.

Purpose of the Study:

• To develop and implement ML algorithms for predicting anti-leishmanial activity of molecules.
• To identify potential drug candidates from existing FDA-approved drugs against Leishmania major.
• To elucidate potential molecular targets for identified anti-leishmanial compounds.

Main Methods:

• Calculated five molecular fingerprints (FPs) for 65,057 molecules using the RDKit library.
• Trained and evaluated four ML algorithms (Random Forest, SVM, etc.) for classifying molecules as active or inactive against Leishmania major.
• Performed virtual screening of FDA-approved drugs and reverse docking against Leishmania drug targets.

Main Results:

• Random Forest and Support Vector Machine models demonstrated superior performance, with atom-pair and topology torsion FPs as optimal embedding functions.
• The models successfully predicted known anti-leishmanial agents among FDA-approved drugs, identifying seven agents within the top 10 predictions.
• Reverse docking experiments identified potential molecular targets for four of the predicted anti-leishmanial drugs.

Conclusions:

• The implemented ML approach, utilizing specific molecular fingerprints and algorithms, is robust for predicting anti-leishmanial activity.
• This study validates the utility of ML in repurposing existing drugs for leishmaniasis treatment.
• The findings provide novel insights into potential anti-leishmanial compounds and their mechanisms of action.