기사 제공처 : The Young Press / 등록기자: 임다혁 [기자에게 문의하기]  0 /  

해당 기사에 관련하여 문의하기에 남겨주시면 "임다혁"기자에게 전송됩니다

 이름

 연락처

- -

 이메일

 확인

Algorithms to Molecules: AI's influence on Drug discovery

By: Leo Dahyuck Im

In July 2026, rentosertib entered Phase III clinical development for idiopathic pulmonary fibrosis, or IPF. The drug is not approved and the Phase III trial has not yet shown whether it works. However, its progress is significant in that artificial intelligence helped identify its target, TNIK, and helped design the small molecule that acts on it. Rentosertib is therefore a useful example of what AI can contribute to drug discovery as well as what it still cannot do by itself.

Developing a medicine has always been a long and tardy process, though it is a mistake to describe traditional drug discovery as simply testing millions of existing chemicals until one happens to work. Researchers have used medicinal chemistry, structural biology, computational models, and repeated rounds of design and testing for many years, with large scale screening is one part of this process, rather than being the entirety of it. Scientists still have to identify a useful biological target, design or improve a compound, and test whether it is active selective, safe, stable, and practical to manufacture. 

AI changes the process by helping researchers work through these decisions more quickly and on a larger scale. Machine learning systems can analyze biological and chemical data, rank possible disease targets, predict some properties of molecules, and generate new molecular structures. This does not mean that AI invented the idea of designing a drug for a specific target. Instead, it gives scientists an alternative to search for targets and compounds and to decide which possibilities deserve the testing first. 

AI and Molecular Modeling:A digital visualization demonstrating how machine learning software renders 3D chemical structures and evaluates small molecule interactions. (Image courtesy of Editorial Stock)

One of the recent examples is Rentosertib, formerly known as ISM001-055, an AI generated small molecule inhibitor being studied for idiopathic pulmonary fibrosis, or IPF. IPF is a progressive lung disease in which scar tissue builds up in the lungs, making it greatly difficult for patients to breathe. In this study, generative AI was put into use to identify TNIK as a possible disease target and to help design a molecule capable of acting on that target. 

Rentosertib, formerly known as ISM001-055 or INS018_055, is an oral small molecule inhibitor being studied for IPF. IPF causes progressive scarring in the lungs, which makes it difficult for patients to breathe. In this program, AI was used in two related but rather distinct ways. Insilico's biology platform prioritized TNIK as a potential target involved in fibrosis. Its generative chemistry platform then assisted the production and optimization of a molecule designed to inhibit TNIK.

Researchers then tested TNIK inhibition in laboratory experiments and in animal models to see whether it actually affected the fibrosis related processes. The candidate then moved into Phase I studies in healthy participants, where researchers examined safety, tolerability, and pharmacokinetics. Those steps were necessary because a promising prediction is not equivalent to a functioning medicine. 

The major test that came next was the GENESIS-IPF Phase IIa trial. The test consisted of a randomized, double blind, placebo controlled study involving 71 patients at 22 sites in China. Patients received either placebo or one of three rentosertib doses for 12 weeks. The primary endpoint was safety and tolerability, and the rates of treatment - emergent adverse events were similar throughout the treatment groups. The group which received a 60 milligram dose once daily also showed a mean forced vital capacity change of +98.4 milliliters, compared to that of -20.3 milliliters in the placebo group. 

However, these results must be interpreted with caution. The study was small in size, lasted only 12 weeks, and the lung function result was a secondary outcome rather than the main safety endpoint. The authors concluded that the findings support further investigation in larger and longer trials while they didn't establish that rentosertib is an effective treatment. 

Insilico announced the start of a Phase III program on July 2026. According to the company's announcement, the planned trial is randomized, double blind, and placebo controlled with about 320 patients and a treatment period of 52 weeks. This is an important milestone despite it still being a test. Rentosertib thus remains investigational and has not been proven effective or approved for patients. 

The rentosertib program shows how AI could speed up the early part of drug development. Insilico and the Nature Biotechnology paper report that the program took about 18 months from target discovery to nomination of a preclinical candidate. AI may help reduce the time spent sorting through biological data, choosing targets, generating possible structures, and deciding which compounds should be made and tested.

Still, the speed of the early discovery process does not automatically shorten every later stage. Clinical trials still need human participants, control groups, safety monitoring reliable measurements, and enough patients to make the results meaningful. Researchers also need time to observe side effects and determine whether a drug's benefits last. Regulatory authorities must review the evidence before a medicine can be approved as well. 

This is where AI differs from a replacement for the drug development system. An algorithm can suggest that TNIK is work investigating or that a certain molecule may have useful properties. It cannot by itself prove that TNIK is the right target in every patient, that the molecule will be safe in the human body, or that it will improve patients' lives. Those questions can only be answered through experiments and clinical trials. 

Rentosertib Development Pipeline: A status summary explaining the clinical trial tracking parameters for investigational therapies. Data points such as intervention type, condition, and phase status help researchers monitor the progress of AI designed candidates like rentosertib. (Image courtesy of the Pulmonary Fibrosis Foundation Clinical Trials Pipeline)

Rentosertib has recently attracted attention because it connects several parts of the AI drug discovery process in one single program. AI helped prioritize the target, generative chemistry helped design the molecule, and researchers then tested the candidate in laboratory models, animals, healthy volunteers, and patients with IPF. That sequence makes rentosertib more informative than a molecule generated by a computer that has never been tested outside a laboratory.

It is also a useful reminder that the most interesting question is not whether AI can produce a chemical structure, since it most clearly can. The harder question is whether AI can consistently help produce medicines that are safe and effective in real patients. So far, rentosertib provides evidence that AI can contribute to the front end of drug discovery and help a program move forward. It does not yet show that AI can guarantee success in clinical development.'

For now, the conclusion is that AI is becoming a powerful research tool rather than an independent inventor of medicines. Rentosertib's progress suggests that AI can help scientists find targets and design molecules faster, but the rest of the process still depends on conventional laboratory work, clinical judgement, patient participation and regulatory review. The Phase III trial will provide a much stronger test of whether the early promise of this approach can translate into a meaningful treatment for IPF.