When Alexander Fleming noticed mould was preventing bacterial growth in a lab dish, he discovered that the fungal mould naturally secreted a biochemical called penicillin. That natural product, now tweaked for commercial-scale delivery, remains in use as an antibiotic to this day.
Alongside colleagues from the A*STAR Institute of Advanced Intelligence and Computing (A*STAR IAIC) and the A*STAR Institute of Molecular and Cell Biology (A*STAR IMCB), the Specialty Chemicals & Bio-Technologies (SCBT) team from the A*STAR Institute of Sustainability for Chemicals, Energy and Environment (A*STAR ISCE2) turned to artificial intelligence (AI) to scour nature’s vast chemical landscape faster and more precisely.
One challenging aspect is that a microbe might carry the genomic code for a biochemical, but not the metabolic machinery needed to produce it—akin to having a recipe book without the cooking equipment. As such, the researchers designed their framework to look for a matching signal coming from both components, biological information from the Protein Language Model (PLM) and predicted chemical information from the Workflow for Intelligent Structural Elucidation (WISE).
To uncover hidden chemical recipes, the PLM reads individual, scattered protein sequences from the genome, instead of looking for exact DNA matches. This enables finding microbial producers that traditional tools might miss when the available genomic data are fragmented or incomplete. Next, WISE leverages generative AI to simulate new chemical structures based on existing knowledge about how other natural products are made. By gaining insight into the potential functional benefits of these unknown compounds, researchers can rapidly prioritise high-value targets for laboratory testing.
The matching of PLM and WISE outputs now enables scientists to discover promising candidates with 75 to 100 percent precision. “Our multi-modal integration framework can be thought of as a high-tech matchmaking system that identifies which microbe can produce a specific, valuable chemical,” said Tay.
Applying their framework to over 2,000 samples, the researchers narrowed the search for microbial producers of a commonly used gram-negative bacteria targeting antibiotic, neomycin B, down to just four top candidates. This targeted strategy drastically reduced screening and successfully yielded two validated species that were previously uncharacterised.
Besides identifying candidate microbial producers, Tay explained that the team hopes to expand the framework to directly support sustainable manufacturing of industrially valuable specialty chemicals. They plan to leverage smart, autonomous AI agents to build a platform that can evaluate chemical, enzymatic and biological pathways simultaneously to design more optimal production pipelines.
“With our team's multi-disciplinary expertise, we hope to develop a platform that can design hybrid, step-by-step synthesis pathways to assign the absolute best, most sustainable technology for each stage of biomanufacturing,” said Tay.
The A*STAR-affiliated researchers contributing to this research are from the A*STAR Institute of Advanced Intelligence and Computing (A*STAR IAIC), A*STAR Institute of Molecular and Cell Biology (A*STAR IMCB), A*STAR Institute of Sustainability for Chemicals, Energy and Environment (A*STAR ISCE2) and A*STAR Bioinformatics Institute (A*STAR BII).
