Using Artificial Intelligence to Optimise Small-Molecule Drug Design
Here at BenevolentAI, we are unlocking the power of scientific data so no disease goes untreated. We achieve this in a variety of ways, right from initial target identification and validation, through hit discovery and lead optimisation, and finally into the clinic, making us the only end-to-end drug discovery company driven by Artificial Intelligence (AI) in the world.
We power this innovation by using structured and unstructured data sources to learn new insights and relationships at scale that otherwise would not be possible. Here, a key differentiator is how we use AI to extract the knowledge locked in the scientific literature and patents to boost our knowledge graph of entity relationships of genes, targets, molecules, and diseases.
My role at BenevolentAI, as head of the Chemoinformatics team, is to guide the scientific direction and validity in the development of our molecular design platform. The size of drug-like chemistry space is truly vast. As an analogy, if we take six Lego bricks, it is possible to construct them in almost one billion unique configurations. Replacing the bricks for atoms, and scaling the number up from six to the more typical size of a drug-like molecule of 20-30 heavy atoms, the size of the space expands dramatically to truly astronomical proportions. The size of this space makes it technically challenging to exhaustively examine every theoretical molecule, instead we use advanced AI algorithms to effectively sample that space to explore and exploit the most promising candidates to take to synthesis and testing.
Drug discovery itself is an inherently multiobjective optimisation process, with many different parameters needed to be optimised in concert. We score each of the candidate solutions with multiple predictive models using a range of appropriate parameters, including the introduction of synthetic tractability, and even planning synthetic routes.
The platform we have developed at BenevolentAI gives us the power to not only tell our scientists what to make, but also how to make the molecules that are of most relevance to our drug discovery programmes, thereby helping to optimise the whole of our AI-driven drug discovery pipeline.
Check out Nathan’s latest publication