What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Yes—AI can help identify biological targets and design candidate molecules, and one such candidate, rentosertib, has reached a randomized phase 2a trial. That is evidence that generative biology can contribute to drug discovery, not proof that AI has independently produced a safe, effective or approved medicine.
What generative biology does in drug discovery
Generative biology applies computational models to biological data to propose things researchers can investigate, such as a disease-related target or a molecule designed to act on it. A target is a biological component—often a protein or pathway—that a treatment aims to affect. A candidate molecule is a proposed compound intended to interact with that target.
The model’s output is a hypothesis, not a result. Researchers still need to make or obtain the candidate, test what it does in biological systems, assess safety, and determine whether it benefits people in clinical studies. Generative models may contribute to parts of this process; that does not mean every stage is automated or that a model alone establishes a treatment’s value.
Free tools Windows power users keep installed
One-click scans. No signup required.
From proposal to evidence
- Nominate a target: A model may help identify a biological target associated with a disease. The proposed relationship needs experimental support; a model’s suggestion by itself does not establish that changing the target will help patients.
- Design a candidate: A model can propose molecules intended to act on a target. Those candidates must be synthesized or otherwise produced and tested to see whether they behave as intended.
- Evaluate in people: Clinical trials examine safety and potential benefit in participants. The design, size, duration and measured outcomes determine what a trial can—and cannot—show.
Each step raises a different question: Did the model propose something useful? Does the candidate act on the intended biology? Does that effect translate into a meaningful clinical benefit, with acceptable risks? Evidence at one step does not answer all the others.
#1 Best Overall
What the rentosertib trial showed
Rentosertib, previously called ISM001-055, is a generative-AI-discovered TNIK inhibitor studied for idiopathic pulmonary fibrosis (IPF). The authors of a Nature Medicine report describe it as the first phase 2a trial of a molecule generated using AI alongside a target discovered using generative AI. The paper describes a pipeline involving AI contributions to target discovery and small-molecule design; it does not establish that AI performed every stage or was solely responsible for the clinical findings.
Study design and safety findings
The randomized, double-blind, placebo-controlled phase 2a trial took place at 21 sites in China. It assigned 71 participants to three rentosertib dose schedules or placebo and followed them for 12 weeks; 55 participants completed that period. The primary endpoint was the proportion of participants with at least one treatment-emergent adverse event. The reported proportions ranged from 70.6% with placebo to 83.3% in two active-treatment groups. The paper also reports that liver toxicity and diarrhea were among events leading to treatment discontinuation.
Rank #2
An exploratory lung-function signal
Forced vital capacity (FVC), a measure of how much air a person can forcibly exhale, was one of the reported outcomes. In the highest once-daily dose group, the mean FVC change was +98.4 ml (95% confidence interval, 10.9 to 185.9), compared with −20.3 ml for placebo (95% confidence interval, −116.1 to 75.6). These estimates come from a small, 12-week study. They are an exploratory signal, not proof that rentosertib improves IPF or that the observed difference will hold in a larger trial.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsThe authors identify small group sizes, short follow-up, geographic and demographic homogeneity, and withdrawals as limitations. They say larger, longer studies are needed to evaluate efficacy and safety. The trial therefore demonstrates that a generative-AI-associated candidate was tested in people; it does not establish a proven treatment or long-term safety.
Rank #3
Does AI make drug discovery faster?
AI may help researchers search biological data and generate candidate designs, but a faster computational step does not by itself prove that the full drug-development process is faster. A candidate still has to clear experimental and clinical tests. The rentosertib paper includes development-timeline claims by its authors; those claims should be understood as specific to the reported program, not as evidence of a typical speed advantage across drug development.
FDA reported that it had received more than 500 drug and biological product submissions containing AI components since 2016. That is a count of submissions with AI components—not a count of generative-biology drugs, successful AI discoveries or approved AI-designed medicines. FDA Commissioner Robert M. Califf 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 particular candidate is safe or effective. See the FDA announcement.
Rank #4
How regulators assess AI-supported evidence
FDA’s January 2025 draft guidance proposes a risk-based approach to assessing the credibility of AI models used to produce information or data intended to support regulatory decisions about the safety, effectiveness or quality of drugs and biological products. Its central idea is to judge a model in relation to its specific context of use: what it is being used to do and how much the regulatory decision depends on its output. The document is draft guidance and expressly nonbinding; it is not an endorsement of a particular platform, candidate or discovery claim. Read the FDA draft guidance.
For engineered biological products, FDA also has final guidance on human gene-therapy products incorporating genome editing. It addresses information for investigational new drug applications, including product design, manufacturing and testing, nonclinical safety assessment, and clinical-trial design. That is regulatory background for genome-editing gene therapies, not guidance specifically about generative AI. See the FDA genome-editing guidance.
Best Value
How to judge claims about an AI-discovered medicine
The label “AI-discovered” covers a range of contributions. To understand what a claim actually means, look for evidence at each step rather than treating the label as a verdict:
- What did the model propose? Was it a biological target, a molecule, or both?
- Was the candidate made and tested? A generated design is not the same as an experimentally validated compound.
- Was the target’s role supported? A proposed target needs biological evidence, not just a computational association.
- What clinical evidence exists? Check the trial phase, comparison group, sample size, follow-up period and outcome measures.
- What remains uncertain? Early studies may leave safety, efficacy, durability and reproducibility unresolved.
Comparisons of development timelines are especially easy to overstate: company-reported timelines do not establish how the same program would have progressed without AI. The stronger test is whether a proposed candidate produces reproducible experimental results and ultimately shows meaningful benefit with acceptable risk in well-designed 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.
