By Nidhi Malhotra
2025 is shaping up to be the year artificial intelligence makes its mark in human biology. The world of drug discovery, long characterised by years of painstaking trial-and-error, is undergoing a seismic transformation. Recent research led by Alex Zhavoronkov, founder and CEO of Insilico Medicine, a Hong Kong-based biotech company, has resulted in a breakthrough: the first drug fully generated by artificial intelligence has entered Phase IIa clinical trials.
The drug, Rentosertib (ISM001-055), is designed to treat idiopathic pulmonary fibrosis (IPF), a fatal lung disease with limited treatment options. With the drug showing encouraging efficacy and safety in human patients, all eyes are turning toward Phase III, the critical next step in trials that could mark the first time a drug entirely generated by AI receives regulatory approval.
Historically, drug development has been a long, expensive process, and fraught with failure. The path from molecular idea to approved therapy takes an average of 10 to 15 years and costs more than $2 billion (Rs 14,547.5 crore).
The process starts with target identification —finding a gene, protein, or pathway in the human body that plays a central role in the disease. This may seem simple in theory, but target identification is one of the toughest hurdles in drug discovery. Biological systems are complex and unpredictable, and oftentimes, attacking seemingly promising targets fail to produce therapeutic benefits or cause harmful side effects in humans. Picking the wrong target early on can sink an entire drug development program before it ever reaches the clinic.
Once a target is validated, scientists search through millions of molecules to find ones that bind to it like keys fitting into locks. These “hit” molecules are then refined into “lead” compounds with the right balance of potency, selectivity, and safety. Only the best of the best are sent into preclinical trials, traditionally in cells and animals to test toxicity and biological activity, though recent regulations now allow advanced lab-based models to partially replace animal testing.
The few that survive this gauntlet move on to human clinical trials, which unfold in carefully staged phases: Phase I to test basic safety, Phase II to explore efficacy in patients, Phase III to confirm effectiveness on a larger scale, and finally Phase IV, which monitors long-term safety after approval. Very few drugs survive this entire journey. For decades, this has meant that pharmaceutical innovation moved at a glacial pace.
What sets AI-driven approach apart
What distinguishes Alex Zhavoronkov’s approach is its seamless integration of AI-powered target discovery and ‘hit’ molecule generation into a closed-loop pipeline – a first in the industry. The team leveraged PandaOmics to analyze complex biological data and identified a protein TNIK (Traf2 and NCK-interacting kinase) as a target, a novel fibrosis driver discovered entirely through AI.
TNIK had been previously studied in cancer, but its role in fibrosis opened a fresh therapeutic pathway. They then employed Chemistry42, the platform’s generative chemistry engine, where 30 AI models worked in parallel, sharing feedback and efficacy scores in real time to explore the enormous universe of possible molecules and refine compounds.
The process yielded Rentosertib, a small molecule with optimised potency, selectivity, and drug-like properties, specifically targeting TNIK. Unlike earlier AI tools that were bolted onto isolated steps of discovery, this end-to-end framework enabled them to go from target discovery to a preclinical candidate in about 18 months, and complete phase 0/1 clinical testing in less than 30 months, dramatically faster than traditional drug discovery.
The therapy has so far demonstrated both safety and signs of efficacy, making it the first AI-discovered and AI-designed drug to clear this bar, something previous AI approaches never achieved.
Insilico is not alone in this revolution. A wave of other pioneering companies is also leveraging AI to transform the drug discovery landscape, with a growing number of candidates advancing to and through the clinical stage.
For example, Atomwise, which uses deep learning for structure-based drug design, announced its first AI-driven development candidate, a medicine that blocks the activity of an enzyme called TYK2 which plays a key role in causing inflammation, in 2023. This is progressing toward clinical trials for immune-mediated inflammatory diseases, a group of chronic conditions where a person’s immune system becomes overactive causing it to mistakenly attack the body’s own healthy tissues.
Not every AI developed drug works. Recursion Pharmaceuticals, which uses AI to generate large biological datasets, had to strategically discontinue its lead AI-discovered candidate, REC-994, for a rare neurovascular disease in May 2025 after long-term data did not confirm earlier efficacy trends. However, the company, having merged with Exscientia in 2024, still has a pipeline of other AI-driven candidates.
While AI is accelerating drug discovery, the complexities of clinical efficacy remain a significant challenge to be addressed as the field matures.
AI’s success rates
A 2024 analysis in Drug Discovery Today found that AI-designed molecules boast early-phase trial success rates well above the industry norm. In Phase I trials, where safety is the main focus, AI drugs achieved success rates between 80 and 90 percent, compared to the historical average of 40–65 percent. In Phase II, where the bar shifts to efficacy, AI-derived drugs have success rates of around 40 percent, at par with traditional methods. Though the sample size remains small, this early evidence suggests that AI is at least as good, and possibly more efficient, at identifying viable drug candidates. More notably, integrating AI across the pipeline nearly doubles R&D productivity.
While Rentosertib marks a historic milestone as the first fully AI-designed drug to reach Phase IIa trials, challenges remain. Current data are from early-stage studies with a limited number of patients, and long-term efficacy and safety across broader populations are yet to be established.
Nevertheless, AI’s ability to combine target discovery, molecule design, and optimization into a single workflow has already accelerated drug development significantly. As regulators become more open to innovative technologies, artificial intelligence is moving beyond promise to practice, transforming how clinical trials are planned, executed, and interpreted, and heralding a new era in pharmaceutical research.
Dr Nidhi Malhotra is an Assistant Professor in the Department of Chemistry at Shiv Nadar Institution of Eminence, Delhi-NCR, India.
