OpenAI Codex and ChatGPT accelerate antimicrobial molecule discovery

TL;DR

OpenAI announced that researchers are leveraging Codex and ChatGPT to accelerate the early-stage discovery of antimicrobial molecules, cutting the time to generate candidate lists from years to hours.

AI‑driven search of biological “code” shortens discovery timelines

The lab led by bioengineer César de la Fuente treats DNA and protein sequences as an information system, training deep‑learning models to recognize functional patterns across vast genomic and proteomic databases. By scanning these datasets with AI, the team can prioritize a manageable set of candidate molecules for experimental validation in hours rather than the multi‑year timelines typical of traditional discovery pipelines.

Integrating Codex and ChatGPT across the research workflow

  • Hypothesis generation – ChatGPT serves as a brainstorming partner, helping the team formulate and refine research questions.
  • Code creation and refinement – Codex translates natural‑language prompts into executable scripts, enabling biologists with limited programming experience to process and analyze large datasets.
  • Data handling – AI tools download, organize, and pre‑process genome and protein data, reducing manual data‑curation effort.
  • Cross‑disciplinary communication – ChatGPT clarifies terminology, summarizes methods from other fields, and even translates content into researchers’ native languages, lowering barriers between biology, chemistry, computer science, and engineering.

“Codex and ChatGPT help bridge those gaps, allowing biologists to build programs and programmers to tackle biological problems.” – César de la Fuente

From AI predictions to experimental validation

AI‑identified candidates must still pass a rigorous experimental pipeline:

  1. In‑vitro efficacy – Confirm that the molecule kills the target microbe and determine the minimal effective concentration.
  2. Cytotoxicity – Assess toxicity to human cells.
  3. Medicinal chemistry optimization – Improve potency, safety, and stability.
  4. Pharmacokinetics & resistance profiling – Evaluate dosing, distribution, and the likelihood of resistance development.
  5. Manufacturability and regulatory review – Ensure scalable production and compliance before clinical trials.

De la Fuente stresses that AI predictions are only useful when paired with “ground‑truth experiments,” emphasizing the necessity of laboratory validation at every stage.

Transdisciplinary collaboration enabled by AI

The lab’s composition spans biology, chemistry, computer science, and engineering. AI tools lower the expertise threshold, allowing members to:

  • Review unfamiliar literature quickly.
  • Generate and test code without deep programming training.
  • Share ideas in a shared ChatGPT workspace that aggregates diverse perspectives.

“Our ChatGPT workspace is receiving input from all these different people that think differently about the problems that we’re trying to tackle.” – César de la Fuente

Limitations and safeguards

While AI accelerates hypothesis generation and data processing, the researchers caution that all outputs must be double‑checked for accuracy. Over‑reliance on AI without experimental verification could lead to false leads.

Broader significance

By treating biological sequences as an alphabet and applying large language models to decode functional patterns, the approach demonstrates a new paradigm for drug discovery: AI‑first candidate screening followed by targeted wet‑lab validation. If widely adopted, this workflow could dramatically increase the pace at which novel antimicrobials reach the testing pipeline, addressing the urgent global threat of antimicrobial resistance.

Outlook

De la Fuente frames AI as the latest tool in a long lineage of scientific instruments—from telescopes to microscopes—that expand human insight. Continued co‑evolution of AI models and experimental biology is expected to deepen our understanding of the “code of life” and unlock therapeutic possibilities that were previously inaccessible.

Sources