, ,

AI in Drug Development: From Protein Models to the First Clinical Trial

Abstrakte Grafik: Molekülstruktur und Kapsel auf hellblauem Grund

Developing a new medicine is lengthy, expensive and risky: it often takes more than ten years from the first idea to approval, and most drug candidates fail in clinical trials. Artificial intelligence is intended to help at several points in this process – and is now delivering its first measurable results.

The breakthrough in protein structures

Many medicines work by binding to a specific protein and altering its function. This requires knowledge of the protein’s three-dimensional shape. For a long time, determining it was laborious and often took years. The AI system AlphaFold made it possible to predict protein structures with high accuracy from the sequence of their building blocks. For this work and for the computational design of new proteins, the 2024 Nobel Prize in Chemistry was awarded to Demis Hassabis, John Jumper and David Baker.

Where AI helps in the development process

  • Finding targets: AI searches genetic, molecular and clinical data for proteins that are involved in a disease and are suitable as drug targets.
  • Designing molecules: generative models propose new chemical structures that could bind to the target while also being well tolerated.
  • Predicting properties: solubility, absorption in the body, possible toxicity – much can be estimated in advance on the computer, so that fewer candidates fail in the laboratory.
  • Planning trials: AI can help to find suitable trial participants or identify subgroups that benefit particularly.

A first clinical milestone

In June 2025, the company Insilico Medicine published in Nature Medicine the results of a phase 2a trial of the drug candidate rentosertib in idiopathic pulmonary fibrosis, a progressive scarring of the lungs. Both the target (the enzyme TNIK) and the molecule itself were identified or designed with the help of AI.

  • 71 patients were treated for twelve weeks.
  • At the highest dose (60 mg once daily), forced vital capacity, a measure of lung function, increased by an average of 98.4 ml. In the placebo group it fell by 20.3 ml.

This is an encouraging signal – but not yet proof of efficacy. That is to be provided by a phase 3 trial with around 320 patients in China, which Insilico started in summer 2026.

Realistic expectations

AI-designed drugs must also pass through all phases of clinical testing. Human biology remains complex, and many reasons why medicines fail – such as unexpected side effects or a lack of efficacy in the disease – can be predicted on the computer only to a limited extent. Even leading companies have had to adjust their timelines: Isomorphic Labs, a spin-off of Google DeepMind, announced in January 2026 that it aimed to start its first clinical trials by the end of 2026 – later than originally planned.

The greatest benefit may therefore initially lie in faster and cheaper early-stage research. Whether AI will also significantly increase the success rate in clinical trials will only become clear in the coming years.

Conclusion

AI has already fundamentally changed basic research on proteins and has brought the first drug candidates into the clinic. However, before AI-designed medicines reach the pharmacy shelf, large trials and regulatory reviews are still needed – the same ones that apply to every other medicine.

More on this topic: AI in Medicine 2026 – an overview.

Note: This article is for general information only. The drugs mentioned are in clinical trials and are not approved. As of: October 2026.

Video on this article