AI in Drug Discovery


A couple of months ago I gave a presentation to the Digital Leadership Forum, an organisation that brings together companies large and small to discuss a range of digital topics.

In the year and a bit since myself and Ben founded Crossr, this has really been the first time I've presented to an audience outside of a pitch capacity and as such the presentation was an opportunity to explain our thesis on where biology, AI, graph computation and software collide.

I'd love the thesis to evolve over time and as such invite you to comment on my thoughts here. You can also watch my presentation below.

TL;DR of our thesis

  • The biopharma industry is a source of incredible human innovation and we are entering a world where every disease target will be druggable
  • The "Better than the Beatles" problem means it's getting harder to create new drugs, increasing drug quality through picking better drug targets is the best way to counteract this
  • Computational and AI approaches based around knoweldge graphs that represent biological networks are complementary to traditional scientific methods, allowing gene/protein/disease characterization based on a much broader set of data than possible through narrow scope hypothesis based analysis
  • Main use cases in drug discovery include target identification and validation, drug repurposing and also broad application recommendation systems for prioritisation of genes, proteins, diseases and drugs
  • However it's challenging to build high quality datasets that represent biological systems as well as gain organisational trust in them due to high dimensionality, high inter-dependency and low stability of biomedical information
  • Software and Operating Systems will increase the quality and speed of science conducted as well as clarity of enterprise decision making by providing a transparent view of data provenance, allowing for real-time data ingestion, querying and advanced analytics and supporting rapid communication of insights across teams

"The challenge with biomedical data availability is often turning data repositories into trusted analytics-grade datasets, there's a lot of work involved in that process"

A panel discussion followed which touched on key areas such as how pharma and startups should collaborate, biomedical data availability and investor perspectives.