Searching for drug candidates with AI

Contents

The development of new drugs is a resource- and time-intensive process that largely leads to a dead end: around 90 percent of development projects are unsuccessful. For years, the pharmaceutical industry has therefore spoken of the great potential of AI tools to massively accelerate drug development. So far, however, there has not been much to show for it. That is set to change soon.

According to a report by the Wall Street Journal, several major pharmaceutical companies are already working at full speed on the development of supercomputers to help discover and develop new drugs.

The pharmaceutical industry and Nvidia

In October last year, US group Eli Lilly, known for its insulin preparations, announced a partnership with Nvidia. The stated goal is to develop the most powerful supercomputer in the industry. Components of the current Vera Rubin GPU architecture are to be used.

Last January, the partnership was expanded to include a five-year collaboration worth up to one billion US dollars. Employees from both companies are already working with AI tools in a newly created innovation lab in Silicon Valley to search for new drug candidates.

The pharmaceutical company Roche has been working with Nvidia since 2023. Last March, the Swiss company announced that it would expand its existing collaboration with more than 3,500 Nvidia Blackwell GPUs – the largest GPU cluster of a pharmaceutical company to date. The Nvidia GPUs are used in hybrid cloud and local environments in the US and Europe. They are intended to train AI models specialized in pharmaceutical applications and to support drug development at individual sites.

No major breakthrough

Despite individual successes with AI technology, the hoped-for groundbreaking successes in drug research have so far failed to materialize. This is partly because the amount of scientific training data for medical AI models is limited, says Najat Khan, CEO of the US biotechnology company Recursion Pharmaceuticals, in the Wall Street Journal report. In addition, there are the high costs of AI-assisted simulations.

Recursion Pharmaceuticals was founded in 2013 with the goal of training AI models with images of cells. The aim is to make the causes of diseases more understandable. This is expected to reduce the failure rate in drug discovery.

Recursion Pharmaceuticals has already achieved minor successes with its AI platform. With the help of AI, the development time for an experimental cancer drug was reduced from four to one and a half years. AI also helped to find out that a certain protein in the human body could be relevant for a colon polyp disease. However, the major breakthrough has so far failed to materialize: even 13 years after its founding, Recursion Pharmaceuticals has not yet been able to bring an AI-based drug to market.

First successes in Japan

Meanwhile, two Japanese pharmaceutical companies have apparently achieved greater successes. According to the Wall Street Journal, drug manufacturer Astellas is said to have used AI to optimize the drug Setidegrasib. The drug is intended to help treat pancreatic cancer and is currently in the late stages of clinical trials.

Meanwhile, the company Takeda has successfully clinically tested a drug for the treatment of psoriasis and plans to apply for approval from the Food and Drug Administration (FDA) in 2026. The active ingredient was discovered by the US company Nimbus Therapeutics using AI and subsequently sold to Takeda.

Major US tech companies are also involved: AlphaFold, developed by Google's AI division DeepMind, can predict protein structures using machine learning, which could be useful for drug development. Last April, OpenAI released GPT-Rosalind. The AI model is tailored for biology, drug discovery, and the implementation of research results in healthcare. (rah)