AI drug discovery

Centaur Biologist

Selecting which targets to work on within a chosen therapeutic area is one of the biggest challenges of drug discovery. A steady progression of downstream effort flows from that single decision.

At Exscientia, we have taken the view that no single laboratory can offer the solution to target selection. Instead, we have concentrated on novel approaches to mine the public corpus of genetic data and biology literature to determine when there is sufficient information to initiate work on a target or interest.

Our systems apply deep learning to comprehensive knowledge graphs allowing us to understand and exploit global research trajectories and emerging hotspots of attention.

Druggability and Tractability

Our AI-driven approach to drug discovery has been at the forefront of new approaches to small molecule drug discovery since we first published it in Nature, in 2012. Since that time, we have extended and refined the platform through the application to live projects and continual internal research.

In every project, we begin with a drugability and tractability assessment: this includes the assimilation of all data available and a determination on the opportunity for a particular target to bind a selective, well-balanced small molecule. For novel targets without any supporting data, broader assessments based on target class are made before determining whether to generate seed data through experimental systems such as our in-house biophysical fragment screening.


Most diseases are highly networked, meaning therapies often need to hit multiple nodes to have a sustainable effect. We have extended our AI platform so it can flexibly design and evolve single molecules that work against more than one target.

The most accessible implementation is for dual targets and for these projects we refer to our dual-targeted compounds as bispecific small molecules (not to be confused with bispecific antibodies).

A range of innovative projects become possible with this approach, ranging from selectively hitting two members of the same family to identifying unrelated targets with complementary binding sites. Both approaches have both been exemplified with our partners Dainippon Sumitomo and Sanofi.

Dual Target Success

With Dainippon Sumitomo (DSP) we have delivered a candidate-quality bispecific small molecule for treating psychiatric disease that is being taken forward by DSP. The candidate molecule evolved by Exscientia’s AI system productively interacts with two distinct GPCR targets from different families.

Read the press release

For Sanofi, Exscientia invented a bispecific molecule that interacts with two unrelated targets. In order to identify this opportunity, over 1000 pairs were evaluated for both chemical compatibility and biological relevance. Evolutionary design was implemented on each pair to invent bispecific small molecules and from this total set, the most promising compounds were synthesised and evaluated through pairs of experimental assays.

Read the press release