A deep learning genome-mining strategy for biosynthetic gene cluster prediction

Area of Science:

• Genomics and Bioinformatics
• Natural Product Discovery
• Drug Development

Background:

• Natural products are vital sources of drug candidates, including antimicrobials and anticancer agents.
• These compounds are microbial secondary metabolites synthesized by Biosynthetic Gene Clusters (BGCs).
• Existing BGC prediction algorithms require improved precision and novel class identification capabilities.

Purpose of the Study:

• To introduce DeepBGC, a deep learning strategy for enhanced BGC identification.
• To improve the accuracy and reduce false positive rates in BGC prediction.
• To enhance the ability to identify novel BGC classes and predict their chemical activities.

Main Methods:

• Development of a deep learning strategy named DeepBGC.
• Supplementation with random forest classifiers for BGC product class and activity prediction.
• Application of DeepBGC to analyze bacterial genomes for novel BGCs.

Main Results:

• DeepBGC demonstrates reduced false positive rates in BGC identification.
• The strategy shows improved ability to extrapolate and identify novel BGC classes.
• Previously undetectable putative BGCs were uncovered in bacterial genomes.
• Accurate prediction of BGC product classes and potential chemical activity was achieved.

Conclusions:

• DeepBGC offers superior performance compared to existing machine-learning tools for BGC identification.
• The tool facilitates the discovery of natural products with potentially novel biological activities.
• DeepBGC represents a significant advancement in in-silico BGC identification and natural product research.