Why does drug discovery take so long?
Making a medicine often starts with a simple question: which molecule will block or switch on the right protein in the body? Chemists have to explore a huge number of possible molecules, pick a few, make them, then test them. Each step takes time and money, and most leads go nowhere.
Here the target is DDR1, a protein from the kinase family (proteins that act like switches inside cells). The authors state that it is implicated in fibrosis, the hardening of some tissues, and in other diseases.
Anything that makes the early search faster or cheaper is therefore of great interest, as long as the molecules found really work when tested.
What did the researchers do with GENTRL?
GENTRL is a generative AI model: instead of sorting molecules that already exist, it proposes new structures. It was designed to optimise three things at once: that the molecule can actually be made in a lab, that it is new, and that it acts well on the target.
The team used it to look for potent inhibitors of DDR1, meaning molecules that block this protein. According to the authors, the discovery work took 21 days. The chosen molecules were then made and tested in the laboratory.
This approach relies on reinforcement learning: the AI tries ideas, receives a kind of score according to how well they meet the three criteria, and gradually improves, like a student correcting their own homework.
What were the results?
Four compounds were active in biochemical tests, which are done in a test tube with the protein alone. Two of them were then validated in cell-based tests, which are closer to real life.
Finally, one lead candidate was tested in mice. It showed favourable pharmacokinetics, which means its journey through the body (absorption, circulation, elimination) looked promising for a future medicine.
These results are modest in number, but they show that the molecules suggested by the AI were not only attractive on paper: they also worked in real experiments.
Key figures
- 21 daysto discover DDR1 inhibitors with the AI
- 4compounds active in biochemical tests
- 2compounds validated in cells
What does this change for patients?
For now, nothing for patients: no treatment has been tested in humans. The interest lies elsewhere. The study shows that an AI can quickly suggest realistic molecule leads, which could shorten the earliest phase of drug research.
If the approach is confirmed, it could help explore diseases that lack treatments more quickly and widen the choice of molecules to study.
For people affected by fibrosis, caution is needed: a laboratory lead is not an available treatment.
What are the limits of this study?
The 21 days cover the discovery step. Many years of safety and efficacy testing must follow before a medicine can be given to patients, and most candidates do not make it. Only four compounds were active in the test tube, and a single one was studied in mice.
Results in mice guarantee nothing in humans. This is a first demonstration that opens a path without proving that a medicine will come out of it.
The study also does not compare the AI's approach directly with the traditional methods that chemists use, so we cannot say by how much it really improves on them.
A future medicine, far from your daily life: and Mima?
This study is about drug discovery in the laboratory: no treatment has been tested in people. It has no effect on your care today, and many years would pass before any trial.
If a treatment concerns you, Mima helps you keep your prescriptions and understand in plain words what it does and its side effects. A useful question for your doctor: "Is there a clinical trial I could be considered for?"
Frequently asked questions
What is generative AI in chemistry?
It is an AI that invents new molecules instead of just sorting existing ones. It learns from examples, then proposes new structures.
What is DDR1?
It is a protein from the kinase family that acts like a switch inside cells. The authors state that it is implicated in fibrosis and other diseases.
Does this medicine exist yet?
No. The study stops at laboratory tests and tests in mice. No tests in humans have been done.
Are 21 days enough to create a medicine?
No, they only cover the discovery of candidate molecules. The safety and efficacy steps that follow take much longer.
The original study
Zhavoronkov A, Ivanenkov YA, Aliper A, et al. “Deep learning enables rapid identification of potent DDR1 kinase inhibitors” Nature Biotechnology, 2 September 2019. DOI : 10.1038/s41587-019-0224-x
Read the full study in Nature BiotechnologyThis article summarises a scientific study and is not medical advice.