Researcher employs Codex and ChatGPT to mine genomes for antimicrobial candidates

The research group has integrated the language models Codex and ChatGPT into its workflow for antimicrobial discovery. These models are prompted to interpret genetic sequences and suggest molecules with potential antibiotic properties. This represents a shift from traditional bioinformatics pipelines that rely on rule-based or statistical methods. By using generative AI, the team can explore sequence space in a more open-ended manner.

The approach targets both currently living organisms and extinct species whose genomes are available in public databases. Searching extinct genomes allows the team to access antimicrobial diversity that may have been lost in modern microbes. This broadens the chemical space beyond what is typically screened in contemporary libraries.

Codex and ChatGPT generate hypotheses about which open reading frames or peptide sequences could exhibit antimicrobial activity. The outputs are prioritized based on similarity to known antibiotic motifs or novelty scores. Researchers then synthesize the top candidates for laboratory testing. Experimental validation remains essential to confirm any predicted activity.

Adopting this AI-driven method requires access to sufficient computational resources to run the models at scale. Expertise in prompt engineering and interpreting model outputs is also necessary. The technique does not replace wet-lab work; predictions can yield false positives or miss active compounds that do not resemble known motifs. Consequently, the workflow stops being useful if the generated candidates fail to be synthesized or show no activity in assays.