DeepMalaria: Artificial Intelligence Driven Discovery of Potent Antiplasmodials

Area of Science:

• Computational chemistry and drug discovery
• Artificial intelligence in medicinal chemistry
• Parasitology and infectious diseases

Background:

• Emerging drug resistance in malaria necessitates novel therapeutic strategies.
• Traditional high-throughput screening (HTS) is resource-intensive and time-consuming.
• Existing virtual screening methods face generalization challenges.

Purpose of the Study:

• To develop an artificial intelligence (AI) model for predicting anti-Plasmodium falciparum activity.
• To leverage deep learning for efficient identification of novel antimalarial drug candidates.
• To validate the AI model's performance using diverse compound libraries and phenotypic screening.

Main Methods:

• Developed DeepMalaria, a graph-based deep learning model trained on SMILES strings.
• Utilized a dataset of 13,446 antiplasmodial compounds from GlaxoSmithKline (GSK).
• Employed transfer learning and validated predictions against macrocyclic compounds, approved drugs, and natural products.

Main Results:

• DeepMalaria accurately predicted compounds with nanomolar activity and high inhibition rates (87.5% for >50% inhibition).
• Validated hits included compounds from macrocyclic libraries and approved drugs.
• Identified DC-9237 as a potent inhibitor across all asexual stages of Plasmodium falciparum.

Conclusions:

• DeepMalaria offers an efficient AI-driven alternative to traditional screening for antimalarial drug discovery.
• The model successfully identified promising lead compounds, including DC-9237, a fast-acting inhibitor.
• This approach accelerates the search for novel antimalarial agents to combat drug resistance.