AI-developed pharmaceutical on the brink of final approval: Could technology revolutionize drug development?
In the realm of drug development, Artificial Intelligence (AI) is making significant strides, revolutionising the process by automating time-consuming and costly steps. AI excels at suggesting new molecules to target known proteins or genes, and is efficient at searching and mining large datasets, summarising information, and detecting patterns in data.
However, AI's perspicacity for pitfalls in patients, such as unexpected toxicity, remains less developed. AI-discovered drugs have shown promising success in clinical trials, including Phase 3, but their ultimate approval rates remain limited, though improving. As of now, at least 15 AI-discovered drugs have reached clinical trials, and 5 have received FDA approval.
One source reports an 80%–90% success rate through Phase I trials for AI-developed drugs, much higher than the ~40% for traditional drugs, suggesting better early-phase outcomes. This high success rate may indirectly improve Phase 3 success, as AI contributes to reducing failure rates in clinical trials by approximately 25–30%.
Despite these advancements, Phase 3 clinical trials—the final and most costly hurdle before regulatory approval—still pose major challenges. The traditional high failure rate (around 90% of candidates fail at some stage) remains a barrier. Challenges faced by companies developing AI-discovered drugs include data quality and quantity issues, interpretability (the “black box” problem), regulatory hurdles, compliance with data privacy and ethical regulations, high costs and the need for specialized expertise, and structural limitations of the clinical development pipeline.
One of the most advanced investigational drugs where both the biological target and therapeutic compound were discovered using AI is a small molecule named rentosertib, developed by InSilico Medicine. This molecule, aimed at treating idiopathic pulmonary fibrosis, a lung-scaring condition, is poised to enter phase 3 clinical trials within the next year or two.
Rentosertib was reported safe and well-tolerated in a 71-patient study in China in June. Recursion, another AI-focused biotech, announced last December that it had dosed the first patient with a candidate drug to treat certain solid tumors and lymphoma.
Partnerships between AI biotechs and pharma companies have proliferated, with Isomorphic Labs entering deals with drug firms Eli Lilly and Novartis that could be worth billions. Notably, Recursion, founded in 2013, runs highly automated in-house labs and generates data covering swathes of biology and chemistry before zeroing in on individual diseases.
However, some critics argue that AI relies on data that has been collected, meaning that chemical space has already been explored, and many of the targets are not very novel. AI-focused biotech has hit headwinds, with AI-drug firm Exscientia making substantial staff cuts and narrowing its pipeline in 2024, before Recursion bought it in an all-stock deal.
Medicinal chemist Derek Lowe scrutinised each of the 24 candidates and noted that in almost every case, the targets were already known to be implicated in the disease under investigation. InSilico Medicine needs to show that its candidate effectively treats idiopathic pulmonary fibrosis in large phase 3 trials.
In 2023, Benevolent signed a deal with Merck for $594 million (£439 million). Bridging the gap between AI predictions and regulatory expectations through explainability, standardized data, and integrated development practices is key to pushing more AI-discovered drugs past this final hurdle.
Science has the potential to revolutionize medical-conditions treatments, as demonstrated in the field of drug development. For instance, AI is making significant strides in the health-and-wellness sector, especially in fitness-and-exercise and technology, by automating time-consuming processes and suggesting new molecules. However, the interpretation and prediction of unexpected side-effects or toxicity (a crucial aspect in medical-conditions management) still pose challenges for AI in this domain.
