DeepMind will establish a research laboratory at The Francis Crick Institute, applying machine learning and artificial intelligence techniques to advance understanding of biology, including protein design and genomics.
The multi-year partnership will see a DeepMind research laboratory set up at the Crick to bring together the Crick’s expertise in the study of biology, health and disease with DeepMind’s expertise in artificial intelligence and machine learning.
The work of DeepMind’s team will include building machine learning models to understand and design biological molecules. The Crick’s extensive facilities will allow researchers to perform experiments to confirm the properties of these designs.
The lab will also explore projects in the field of genomics. The researchers will be able to experimentally test the biological hypotheses predicted by their models.
Paul Nurse, director of the Crick says: “We are very pleased to welcome DeepMind into the Crick where expertise in machine learning, high-throughput computing and laboratory experiments can be combined to tackle important biological and biomedical questions.”
Image from DeepMind's Visualising AI project. Here, the application of AI is represented by a ‘wrapper’ which is applied to a series of stylised icons. The icons illustrate the wide range of applications for the technology while the wrap effect roots the concept in a tangible reality. - Tim West
Pushmeet Kohli, Head of AI for Science at DeepMind, says: "The Crick is a world-renowned research institute, whom we are delighted to partner with as we enter a new era in which AI methods can be used to help model all aspects of biological systems, and accelerate new discoveries."
DeepMind is looking to build upon previous successes in the field of biology. In 2020 their system, AlphaFold, was recognised as a solution to the protein folding problem, a grand challenge in biology for the past 50 years. Last year, they released the AlphaFold Protein Structure Database in partnership with EMBL-EBI, containing over one million protein structure predictions including the entire human proteome - more than doubling the number of high-accuracy human protein structures available to researchers. Separately, their neural network architecture, Enformer, has led to increased accuracy in predicting gene expression from different DNA sequences.