A group of researchers from the University of Washington describe a new tool in two papers published in Science.

The papers show how deep learning is changing the way scientists work.

In nature, all the problems of life are solved by proteins. David Baker is the director of the Institute forProtein Design at the University of Washington.

Over the course of evolution, they were able to solve the problems faced by organisms. Covid is one of the new problems we face. It would be really, really powerful if we could design something that solved new problems as well as the ones that evolved during evolution.

Hundreds of thousands of amino acids are linked up in long chains and folded into shapes. Researchers can get insight into how they will behave by using AlphaFold.

The inverse problem will be helped by theProteinMPNN. It will help them find the sequence that folds into that shape if they already have an exact structure in their head. A neural network is trained on a large number of examples of the same sequence folding into three-dimensional structures

Researchers need to find a solution to another problem. The first thing they have to do is figure out what theProtein backbone would do in a real world problem.

The team in Baker's lab uses two machine-learning methods that they call "constrained hallucination" and "in painting."