AlphaFold is an AI system developed by DeepMind that predicts protein 3D structure from its amino-acid sequence. It’s been a year since the AlphaFold software and ‘AlphaFold Protein Structure Database’ were made publicly available for users to explore and investigate their protein of interest.

We are running a webinar series to mark this occasion by highlighting the impact of AlphaFold on training and research in life sciences. This series will comprise 3 webinars:

Overview, scope and vision of AlphaFold - This webinar took place on Tuesday 14 June at 13:30 BST - Click on the 'Watch video' button on this page to listen to the webinar.
How AlphaFold is changing the teaching/training landscape in life sciences - Tuesday 21 June 13:30 BST - Click here to find more information and the recording for this webinar.
Impact of AlphaFold on research and development in life sciences - Tuesday 28 June 13:30 BST - Click here to find more information and the recording for this webinar.

To discuss the overview, scope and vision of AlphaFold we had the following panel of speakers:

Kathryn Tunyasuvunakool - Machine learning models have the potential to become core tools in biology, as recent progress in protein structure prediction illustrates. In this webinar I gave an overview of AlphaFold: how the system works, how to obtain protein structure predictions, and how to analyse them. I then reviewed some ways in which the system has been built upon, and discussed how to evaluate AlphaFold for a new application.

Randy Read - ​​Few areas of structural biology have been untouched by the recent dramatic increases in the power and accuracy of computational modelling of protein structure. These changes have been wrought by the current version of AlphaFold, with RoseTTAFold not far behind. Experimental structural biology is still needed to resolve ambiguities in the predicted structures and to verify the details, but the availability of high-quality models is removing many of the bottlenecks in the experiments. Even without an experimental structure, the new models are sufficient to generate interesting hypotheses that can be tested experimentally, such as assessing how variants associated with genetic disease actually cause disease. Limitations in the models could potentially be addressed by adding explicit physics and chemistry to the pattern recognition used in the current algorithms, and by actively exploiting even limited experimental observations.

Sergey Ovchinnikov - I discussed the impact of AlphaFold on structural bioinformatics by highlighting a few large scale efforts and structure-search tools developed to characterise the AlphaFold models.