The RIght DRUG
The chemical space for exploration is almost infinitely vast and only a small fraction of it can potentially be made into medicines. Medicinal chemists train for decades to recognise and design drug-molecules that ensure safety, potency, duration of impact and novelty. This is a true art form and even the best chemists can only optimise a limited number of properties at any given time.
Accelerated compound optimisation using AI.
By leveraging advanced AI, our EvoChem product is continuously learning from this vast chemical space and generating drug-like molecules with desirable properties that can be synthesised 'on demand'. EvoChem designs de novo compounds based on multiparametric optimisations with a scoring function that factors in all the properties we are seeking to optimise for that molecule. The compound ideas are ranked and the best candidates are selected directly for synthesis, whilst others serve to inspire chemists to further explore the chemical space.
Iterative design cycles combining experimental data and expert feedback.
We then take these selected compounds, synthesise them in our labs and generate real experimental data which is then fed back in, to refine the models. But what’s unique about our process is that it is assisted, from start to finish, by our AI algorithms and generative models.
The result is a system that is self-correcting and requires fewer molecules to be synthesized, speeding up the process of selecting a final candidate from an industry average of 4.5 years to just 14 months.
Benchmarking Models for De Novo Molecular Design.
With the emergence of deep learning and neural generative models, models for molecular design based on neural networks show promising results.
However, the new models have not been profiled on consistent tasks, and comparative studies to well-established algorithms have seldom been performed.
We have released GuacaMol, a framework to benchmark models for de novo molecular design.