A strong baseline sets the bar for peptide screening
A 100-query peptide screening replay reveals a strong fixed ranking baseline and a careful way to test whether feedback can improve on it.
Read the noteBuilding a reinforcement learning platform for biology.
We’re developing a way to use feedback to make better choices across large biological search spaces. Peptide–MHC binding is the first project we’ve initiated to test that approach.
FIRST PROJECT
Starting with peptide–MHC binding.
Our first project asks whether feedback can improve which peptide candidates we choose to examine next. It gives us a defined setting to develop the platform and compare adaptive choices with a strong fixed ranking.
Explore the projectTHE PLATFORM APPROACH
From feedback to the next choice.
We compare adaptive search with simpler methods to test whether feedback changes which candidates we prioritize.
Frame the question
Start with a biological question and define what the available evidence can and cannot tell us.
Explore possibilities
Search large candidate spaces in computation and update priorities as new feedback arrives.
Learn from feedback
Compare adaptive search with simpler methods while keeping predictions distinct from biological results.