Predicting fungal secondary metabolite activity from biosynthetic gene cluster data using machine learning

Area of Science:

• Fungal secondary metabolism
• Bioinformatics
• Machine learning in drug discovery

Background:

• Fungal secondary metabolites (SMs) are crucial for ecological diversity and possess valuable medicinal properties (antifungal, antibacterial, antitumor).
• Biosynthetic gene clusters (BGCs) encode SM production, but predicting SM bioactivity from BGCs is challenging.
• Current data limitations hinder machine learning applications in fungal secondary metabolism.

Approach:

• Adapted machine learning models from bacterial BGCs to predict fungal SM bioactivity.
• Trained models on two datasets: fungal BGCs only and combined fungal and bacterial BGCs.
• Evaluated model performance for predicting antibacterial, antifungal, and cytotoxic/antitumor activities.

Key Points:

• Models trained solely on fungal BGCs achieved 51-68% balanced accuracy.
• Models trained on combined fungal and bacterial BGCs reached 61-74% balanced accuracy.
• Limited bioactivity data for fungal SMs impacts prediction accuracy.

Conclusions:

• Machine learning models integrating bacterial and fungal BGC data show promise for predicting fungal SM bioactivity.
• Increased systematic efforts are needed to link fungal SM bioactivity to BGCs.
• Addressing data scarcity is crucial for advancing machine learning in fungal secondary metabolism research.