BioAsset

Insights / All articles

Research

AI x Drug Discovery: From Computation To Cure

By Dmitry Vysotskiy, Founder & President at BioAsset · 4 slides

The pharma industry faces inefficient economics: an average cost of $2.8B per successful drug, 90% clinical trial failure rates and preclinical timelines of 4-5 years. AI transformer models, the same architecture behind ChatGPT, are now being widely deployed across drug discovery and promise significant changes to those economics.

Recent news

Insilico Medicine pioneered AI-discovered drugs, reducing preclinical development from 4.5 years to 18 months for rentosertib, a drug that treats a scarring lung disease.

Variant Bio launched its “Inference” agentic AI platform and secured a $120M+ deal with Boehringer Ingelheim, a major German pharmaceutical company, for kidney disease drug discovery.

GSK developed Phenformer, an AI tool that links genomics with disease mechanisms, while AstraZeneca reports that 90%+ of its small molecule pipeline is now AI-assisted.

On clinical success rates, AI-discovered molecules are showing 80-90% early trial success, against a historical 40-65%.

The big picture

Data from The Economist and Reuters sets the traditional and AI approaches side by side. The preclinical phase takes 4.5 years the traditional way and 1.5 years with AI assistance, 67% faster. Full drug development falls from 12.0 years to 8.0 years, 33% faster. In early clinical trials, the success rate is 85.0% for AI-discovered drugs against 52.5% for traditional ones, 62% higher. Overall success to market is 13.5% against 7.5%.

The market is being transformed in several ways:

Outlook

OpenAI’s recent entry, partnering with Moderna, and Nvidia’s massive deal with Eli Lilly signal that tech giants see this as the path to AGI in biology. Drug discovery is becoming one of AI’s highest-value applications, and it is increasing demand for CS/ML practitioners with deep scientific literacy.

Companies that successfully integrate AI could see pipeline velocity that is 2-3x faster. Clinical trial risk and assessments remain real, however, and the 80-90% early success rate needs further validation across larger and more diverse trials.

The patent cliff facing big pharma becomes more solvable when pipelines can be filled quicker and cheaper. This could shift M&A dynamics. Expect more acquisitions of AI-discovered late-stage assets and projects.

Real success here depends on whether AI can truly solve and produce novel biological concepts, or whether we are just optimizing within known chemical space more efficiently. Insilico’s rentosertib is proof of the concept, but scaling from 1 successful drug to 40+ will reveal whether this is transformative or just slightly better.

This article is the text version of a BioAsset slide post. See the original slides on bioasset.info.