Over 98% of observed genetic differences in humans do not directly code for proteins. We know that a lot of this type of genetic variation underlies human disease, so deciphering the molecular function of this ‘dark DNA’ is an important frontier in human genetics. Deep learning models are emerging as powerful tools for understanding this regulatory genetic code.

We present AlphaGenome, which takes 1 million letters of DNA sequence as input and predicts thousands of experimental measurements up to single base pair resolution, covering a diverse set of regulatory mechanisms - including gene expression, transcription initiation, chromatin accessibility, histone modifications, transcription factor binding, chromatin contact maps, splice site usage, and splice junction coordinates and strength.

AlphaGenome generalises to predicting the effect of genetic changes (‘variants’) and matches or exceeds the strongest available alternative models on the vast majority of evaluations. AlphaGenome’s ability to simultaneously and accurately predict the effect of variants across multiple molecular phenomena recapitulates the mechanisms of clinically-relevant variants - for example, near the TAL1 oncogene.

To facilitate broader scientific use, we provide an API for making predictions for experimental assays and variant effects from DNA sequence.

This talk will cover an overview of the model, its performance on different tasks, and highlight some important use-cases. We will also outline the practical steps for applying the model in your scientific research.

This is an exciting opportunity to directly talk to and learn from Google researchers building cutting edge genomics models. As our speaker is donating their time to walk us through this pivotal technology we want to ensure an engaged event. Please reserve your spot only if you are committed to joining us for this session.