Generative biology can help researchers propose drug targets and design candidate molecules, but its output is a hypothesis—not a medicine—until experiments and clinical studies test it. One notable milestone is rentosertib, a generative-AI-discovered drug candidate studied in a randomized phase 2a trial for idiopathic pulmonary fibrosis. That small, 12-week study reported an exploratory lung-function signal; it did not establish that the drug works or is safe long term.
What does generative biology do in drug discovery?
Generative biology applies computational models to biological questions. Depending on the project, a model may help researchers identify a possible biological target, propose a molecule intended to act on it, or contribute to both steps. The proposal then has to be made and tested: researchers investigate whether the candidate behaves as expected in experiments, whether the target matters to the disease, and whether the candidate can be studied safely in people.
As an Amazon Associate I earn from qualifying purchases.
This is different from an automated path from prompt to prescription. A generated sequence or compound is not evidence of efficacy, and a model’s suggestion does not by itself demonstrate that a target causes or sustains a disease. The experimental and clinical work remains essential.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Target discovery and molecule design are separate claims
A target is a biological component researchers hope to influence; a candidate molecule is a proposed way to influence it. A system may contribute to one or both tasks. When assessing a claim about an AI-discovered drug, it matters whether the target was experimentally supported, whether the molecule was synthesized and tested, and what stage of human study it reached.
#1 Best Overall
Has an AI-designed drug been tested in people?
Yes. Rentosertib, previously called ISM001-055, is a TNIK inhibitor developed for idiopathic pulmonary fibrosis (IPF). The Nature Medicine trial report describes it as the first phase 2a trial of a molecule generated using AI and a target discovered using generative AI. This is a meaningful clinical milestone, not proof that AI independently created a proven or approved treatment.
What the phase 2a study measured
The randomized, double-blind, placebo-controlled study enrolled 71 participants at 21 sites in China and followed them for 12 weeks; 55 participants completed the period. Its primary endpoint was the proportion of participants with at least one treatment-emergent adverse event (TEAE). The reported FVC changes below are a separate, exploratory measure—not the primary endpoint.
Rank #2
| Study measure | Reported result | How to interpret it |
|---|---|---|
| TEAEs | Rates ranged from 70.6% with placebo to 83.3% in two active-treatment groups | The study’s primary endpoint concerned adverse events, not proof of efficacy. |
| Forced vital capacity (FVC), 60 mg once daily | Mean change: +98.4 ml (95% CI 10.9 to 185.9) | An exploratory result from a small, short study. |
| Forced vital capacity (FVC), placebo | Mean change: −20.3 ml (95% CI −116.1 to 75.6) | The comparison does not establish a clinical benefit on its own. |
FVC measures how much air a person can forcefully exhale after taking a deep breath. The difference in mean changes is a signal worth testing further, but the trial’s size and duration do not justify treating it as conclusive evidence that rentosertib benefits people with IPF.
What remains uncertain
The paper notes small group sizes, short follow-up, geographic and demographic homogeneity, and participant withdrawals. Liver toxicity and diarrhea were among the events leading to discontinuation. The authors say larger, longer studies are needed to assess efficacy and safety. The available evidence here does not establish a later-stage trial or regulatory milestone, so rentosertib should be described as an investigational candidate studied in phase 2a—not as an approved drug.
Rank #3
Does AI make drug discovery faster?
It may help researchers generate or prioritize hypotheses, but the rentosertib trial does not show that generative biology generally makes drug development faster. Drug discovery includes target validation, candidate optimization, laboratory testing, manufacturing, and clinical evaluation. A company timeline or an authors’ account of one development program is not a controlled comparison proving a field-wide speed advantage.
A more useful way to judge a claimed advance is to ask what the model proposed, what experiments tested the proposal, how independently the target’s role was supported, and what clinical evidence followed. Human studies still need enough participants and follow-up to answer questions about benefit and harm.
Rank #4
How are regulators assessing AI in drug development?
In January 2025, the U.S. Food and Drug Administration (FDA) issued draft guidance on AI used to produce information or data intended to support regulatory decisions about the safety, effectiveness, or quality of drugs and biological products. It proposes assessing model credibility according to the model’s particular context of use. The guidance is expressly nonbinding; it is an evidence-assessment framework, not an endorsement of a particular AI platform or candidate. Read the FDA draft guidance.
The FDA said it had experience with more than 500 submissions containing AI components since 2016. That figure counts submissions with AI components; it is not a count of generative-biology drugs or successful AI discoveries. FDA Commissioner Robert M. Califf, M.D., said, “With the appropriate safeguards in place, artificial intelligence has transformative potential to advance clinical research and accelerate medical product development to improve patient care.” The statement describes potential, not a finding that any specific AI-generated candidate is effective.
Best Value
Regulatory documents also cover related but distinct technologies. For example, the FDA’s final guidance on human gene therapy products incorporating human genome editing addresses product design, manufacturing and testing, nonclinical safety assessment, and clinical trial design. It is background for genome-editing products, not AI-specific guidance.
What should you conclude from the current evidence?
Generative biology has moved beyond purely computational proposals: rentosertib reached a randomized phase 2a study, and that study reported an exploratory FVC result alongside safety data. But a candidate reaching a clinical trial is not the same as a treatment being proven, and one example cannot establish that AI routinely finds better drugs or shortens development. The decisive evidence must come from rigorous experiments and adequately sized, longer clinical studies.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
