Drug discovery predictions that say how sure they are
AIChemy is our drug-discovery platform for medicinal chemists. It predicts binding affinity, pKa, logD, and toxicity with calibrated uncertainty, screens compound libraries, and designs new molecules from fragments.
What it does
Most of the cost in early discovery goes into synthesising and assaying compounds that were never going to work. AIChemy narrows the funnel earlier, and tells you which of its calls to trust.
Binding affinity
Predicted affinity of each compound for your target, with an interval, so you can tell the hits the model is sure about from the ones it is guessing on.
pKa and logD
Ionisation and lipophilicity, the properties behind solubility, permeability, and distribution, predicted before anything is synthesised.
Toxicity
Early flags for toxicity liabilities, so problem compounds drop out of the funnel before they reach the lab.
Virtual screening
Rank large compound libraries against a target and spend synthesis and assay capacity on the candidates most likely to pay off.
Fragment-based generative design
Grow and link fragments into new candidate molecules, scored by the same uncertainty-aware models.
Uncertainty on every prediction
Every number comes with a calibrated credible interval. Chemists decide where to trust the model and where to run the experiment.
How it works
AIChemy models molecules as graphs with equivariant deep probabilistic graph neural networks. Equivariance means the models respect the symmetries of 3D chemistry: rotate or translate a molecule and the prediction stays the same.
The models are Bayesian, built on Borch, our open-source probabilistic programming language. That is where the calibrated uncertainty comes from: it is a property of the model, not a score added afterwards.
Read the details in our post on equivariant graph neural networks for drug discovery and the CDK9 inhibitor study presented at DDC Europe 2024.
Built with medicinal chemists
We develop AIChemy together with RG Discovery, who bring the medicinal chemistry and assay expertise that grounds the models in real wet-lab data. We work the same way with pharma and biotech teams: co-development, not a black box.
- 1
Scoping call
Your target, your chemistry, the data you have, and the decision you need to make. 30 minutes with one of our scientists.
- 2
Proof of concept
We run AIChemy on your problem alongside your chemists and measure it against your current pipeline.
- 3
Co-development
If it works, we build it into your discovery workflow together and keep improving the models as your assay data grows.
Funding
The development of AIChemy has received funding from Innovation Fund Denmark and from the Eurostars programme, co-funded by the European Union.
Ready for the next step?
Tell us the decision you need to make with confidence. We'll tell you honestly whether a model can help.