TL;DR
Open a free Amazon Business account
Business pricing, bulk buying and tax-exempt orders.
Create a free accountAs an affiliate, we earn on qualifying purchases.
The University of Pennsylvania lab of bioengineer César de la Fuente uses its own deep-learning models alongside ChatGPT and Codex to search genomes for antimicrobial candidate molecules, reducing an early search stage from years to hours, according to OpenAI. Candidates still face years of lab validation and clinical trials before becoming drugs.
Bioengineer César de la Fuente and his laboratory are using ChatGPT and Codex alongside custom deep-learning models to search genomes for new antimicrobial molecules — an approach the lab says can compress an early-stage candidate search from years to hours, according to an OpenAI report on the lab’s work.
The lab’s core premise is that biology can be treated as an information system. “The nucleotides that make up DNA, and the amino acids that make up proteins and peptides are sort of like an alphabet,” de la Fuente said. His team trains deep-learning models to recognize patterns in biological sequences and scan vast genome and protein datasets for peptides with potential antimicrobial activity.
ChatGPT and Codex play a supporting role rather than a discovery role: lab members use them to brainstorm hypotheses, write and refine code, process datasets, analyze results, and bridge gaps between disciplines. Because the lab spans biology, chemistry, computer science, and engineering, AI tools let biologists build programs and programmers engage with biological questions, de la Fuente said. Some members also use ChatGPT to work in their native languages and to review unfamiliar topics.
The scale of the problem is large. About five million deaths in 2021 were associated with bacterial antimicrobial resistance, a toll projected to roughly double by 2050, according to figures cited in the report. De la Fuente noted that no new class of antibiotics has been introduced in roughly 50 years, and that much modern development modifies existing medicines, an approach with diminishing returns.
How A Researcher Uses Codex And ChatGPT To Search For New Antimicrobial Molecules
Bioengineer César de la Fuente’s lab pairs its own deep-learning models with ChatGPT and Codex to scan genomes for antimicrobial peptides — compressing an early-stage candidate search from years to hours, according to an OpenAI report. Discovery, however, is only step one.
Biology as an Information System
The lab’s core premise: nucleotides and amino acids behave like an alphabet. Deep-learning models trained to recognize patterns in biological sequences scan vast genome and protein datasets — including genomes of extinct organisms — for peptides with potential antimicrobial activity.
Custom Deep-Learning Models
The lab’s own models perform the candidate search — pattern-matching across genomes and protein datasets to surface peptides that may fight microbes.
ChatGPT & Codex
Used to brainstorm hypotheses, write and refine code, process datasets, analyze results, and bridge gaps between disciplines — not to discover molecules.
Lowering Barriers
Biologists build programs; programmers engage with biology. Some members also use ChatGPT to work in native languages and review unfamiliar topics.
From Genome Scan to Clinic — A Long Chain
“Years to hours” covers only the computational search. Every candidate must still clear a long chain of validation before reaching patients.
Genome Scan
Models scan genome and protein databases for candidate antimicrobial peptides.
Activity & Toxicity
Confirm the molecule kills the target microbe; test toxicity to human cells; optimize dosing, safety, and stability.
Resistance & ADME
Test resistance development, how the molecule moves through the body, and manufacturing feasibility.
Trials & Approval
Regulatory review and clinical trials — then, possibly, an approved drug.
From Soil Samples to Genome Databases
Historically, antimicrobials were found by isolating and testing molecules from plants, animals, microbes, insects, water, and soil — a years-long process. Digital databases moved the bottleneck from sample collection to signal detection: only a fraction of any genome has a clearly understood function.
Ground Truth, Double-Checked
“Ground-truth experiments are essential to validate AI predictions.”
— César de la Fuente“Obviously you have to always double-check for accuracy.”
— César de la Fuente“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“They’re essentially at the edges between fields where very few people work.”
— César de la Fuente, on where the promising territory liesWhat the Report Leaves Unverified
The source is OpenAI — which sells ChatGPT and Codex — describing its own products. Here is what the report does and does not establish.
| Claim / Element | Established? | Notes |
|---|---|---|
| “Years to hours” search speedup | ~ Partially | Covers the computational search stage only — not validated drug candidates. |
| Candidates in clinical trials | ✗ Not stated | No numbers given for candidates entering trials or receiving approval; none described as approved drugs. |
| Peer-reviewed results for this workflow | ✗ Absent | The lab has published related peptide work in journals, but the report itself presents no peer-reviewed results. |
| Lab’s pattern-recognition models exist | ✓ Credible | Consistent with previously published work from the de la Fuente lab. |
| Source independence | ~ Interested party | The framing of the tools’ usefulness comes from OpenAI, which sells them. |
Promising Acceleration — Not Yet a Drug
Credible but early-stage
The strongest counterargument: discovery was never the main bottleneck in antibiotics. Toxicology, resistance management, clinical trials, and weak commercial incentives have kept new antibiotic classes off the market for decades — faster candidate generation does not fix those downstream failures, and AI-suggested molecules can carry hidden toxicity or resistance risks that only emerge late in testing.
What would change the assessment: peer-reviewed evidence that a molecule surfaced by this specific workflow has entered clinical trials and shown an acceptable safety profile in humans. Until then — a promising acceleration of the search process, not progress toward an approved drug.
Q.Did ChatGPT discover a new antibiotic?
No. The lab’s own deep-learning models perform the candidate search; ChatGPT and Codex support brainstorming, coding, data processing, and cross-disciplinary work. No approved drug has emerged from the workflow as described.
Q.How fast is the AI-assisted search?
According to the lab, the initial computational search drops from years to hours — covering only early discovery, not validation or testing.
Q.How big is the resistance problem?
About five million deaths in 2021 were associated with bacterial antimicrobial resistance, a toll projected to roughly double by 2050.
Q.Does the lab trust AI outputs unchecked?
No. De la Fuente insists AI and lab biology must advance together, with ground-truth experiments validating AI predictions at every stage: “Obviously you have to always double-check for accuracy.”
Why AI-Accelerated Discovery Matters Now
Antimicrobial resistance is a growing global health threat, and traditional discovery pipelines are slow and expensive. If AI can reliably surface a manageable shortlist of candidate molecules from enormous genomic datasets, it could redirect scarce laboratory time toward the molecules most likely to work.
The lab’s use of general-purpose AI tools such as ChatGPT and Codex also illustrates a broader shift: AI as a cross-disciplinary collaborator that lowers barriers for specialists. “Our ChatGPT workspace is receiving input from all these different people that think differently about the problems that we’re trying to tackle,” de la Fuente said.
However, discovery is only the first step. Each candidate must still clear a long chain of validation before it can reach patients.
molecular biology research software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Soil Samples to Genome Databases
Scientists have historically searched for antimicrobials in plants, animals, microbes, insects, water, and soil, isolating and testing candidate molecules iteratively — a process that can take years. Digital genome and protein databases now allow researchers to search across the tree of life, including genomes of extinct organisms, which de la Fuente’s lab explores.
The bottleneck shifted from sample collection to signal detection: only a fraction of any genome has a clearly understood function, and fewer sequences still encode molecules that can fight microbes. De la Fuente argues the most promising territory lies between disciplines: “They’re essentially at the edges between fields where very few people” work, he said.
“Antimicrobial resistance is one of the greatest existential threats to humanity in my opinion. And yet, we haven’t had a new class of antibiotics for 50 years.”
— César de la Fuente, bioengineer
As an affiliate, we earn on qualifying purchases.
What the Report Leaves Unverified
The “years to hours” claim refers to the computational search stage only, not to validated drug candidates. The report does not state how many candidates found through this pipeline have entered clinical trials or received regulatory approval, and none are described as approved drugs.
The piece is published by OpenAI, which sells ChatGPT and Codex, so the framing of the tools’ usefulness comes from an interested party. De la Fuente’s lab has previously published work on AI-discovered antimicrobial peptides in peer-reviewed journals, but the report itself does not present peer-reviewed results for the workflow described.
antimicrobial peptide discovery kits
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Long Road After Candidate Discovery
Any promising molecule must be confirmed to kill the target microbe, dosed and tested for toxicity to human cells, and optimized by chemists for effectiveness, safety, and stability. Further testing covers resistance development, how the molecule moves through the body, and manufacturing feasibility.
Candidates that clear all of these hurdles still face regulatory review and clinical trials. De la Fuente says AI and laboratory biology must advance together, with ground-truth experiments validating AI predictions at every stage.
As an affiliate, we earn on qualifying purchases.
Where I land
I read this as a credible but early-stage illustration of how AI reshapes the front end of drug discovery. The most solid elements are the lab’s pattern-recognition models and its insistence on ground-truth validation; the weakest are the promotional edges, since the source is OpenAI describing its own products. The concrete claim — candidate search compressed from years to hours — is plausible and consistent with published work from this lab, but it covers only the cheapest part of the pipeline.
The strongest counterargument is that discovery was never the main bottleneck in antibiotics: toxicology, resistance management, clinical trials, and weak commercial incentives have kept new antibiotic classes off the market for decades. Faster candidate generation does not fix those downstream failures, and AI-suggested molecules can carry hidden toxicity or resistance risks that only emerge late in testing.
What would change my assessment is peer-reviewed evidence that a molecule surfaced by this specific workflow has entered clinical trials and shown an acceptable safety profile in humans. Until then, I’d treat this as a promising acceleration of the search process, not as progress toward an approved drug.
Key Questions
Does this mean ChatGPT discovered a new antibiotic?
No. The lab’s own deep-learning models perform the candidate search; ChatGPT and Codex are used for brainstorming, coding, data processing, and cross-disciplinary support. No approved drug has emerged from the workflow as described.
How fast is the AI-assisted search?
According to the lab, the approach can reduce the initial computational search for candidate molecules from years to hours. This covers only the early discovery stage, not validation or testing.
How big is the antimicrobial resistance problem?
About five million deaths in 2021 were associated with bacterial antimicrobial resistance, with the annual toll projected to roughly double by 2050, according to figures cited in the report.
What happens after a candidate molecule is identified?
It must be tested for antimicrobial activity, toxicity, dosing, resistance risk, and behavior in the body, then optimized and manufactured — followed by regulatory review and clinical trials before approval.
Does the lab trust AI outputs without checking them?
No. De la Fuente explicitly cautions, “Obviously you have to always double-check for accuracy,” and stresses that ground-truth experiments are essential to validate AI predictions.
Source: OpenAI
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
