DeepMalaria

DeepMalaria predicts antiplasmodial activity of chemical compounds using deep learning to accelerate identification of novel antimalarial candidates against Plasmodium falciparum.

Key Features:

• AI-Driven Prediction: Uses deep learning on Simplified Molecular Input Line Entry System (SMILES) representations to predict inhibitory properties of compounds against Plasmodium falciparum.
• Graph-Based Model Training: Employs a graph-based deep learning model trained on 13,446 antiplasmodial hit compounds from the GlaxoSmithKline (GSK) dataset.
• Macrocyclic Compound Exploration: Extends predictive coverage to macrocyclic compounds to identify underexplored ligand-binding chemotypes.
• Transfer Learning for Enhanced Performance: Applies transfer learning from a large dataset to improve model generalization across diverse chemical libraries.
• Validation through Phenotypic Screening: Validation includes phenotypic screening using a SYBR Green I fluorescence assay, including detection of nanomolar activity and compounds with greater than 50% inhibition of Plasmodium falciparum.

Scientific Applications:

• Novel antimalarial discovery: Prioritizes novel antimalarial agents against Plasmodium falciparum by predicting compound efficacy.
• Candidate prioritization: Accelerates identification and prioritization of compounds for further experimental testing.
• Drug repurposing evaluation: Assesses approved and existing drugs for potential antiplasmodial activity to support repurposing efforts.

Methodology:

Trains a graph-based deep learning model on SMILES representations of 13,446 antiplasmodial hits from the GSK dataset, applies transfer learning from a larger dataset, predicts activity for novel and repurposed compound libraries, and validates predictions using independent datasets primarily consisting of natural products.