krdav
commented
7 years ago The paper is based on sparse validation data but it seems legit considering that they don't fit on validation data, the method is unsupervised and other papers suggest similar correlations to protein stability and function.
The model is called a Potts model and @dunleavy005 and I have be distracted from our aammp work recently because of it. Here is our paperpile on the topic: https://paperpile.com/shared/SebAmk
Among lots of junk: Ekeberg’s “Improved contact prediction in proteins: using pseudolikelihoods to infer Potts models” is the best in term of describing the method." is a very good paper on the methods and ideas used in nearly all the other models. The “Potts Hamiltonian models of protein co-variation, free energy landscapes, and evolutionary fitness” is a nice and short review. "Coevolutionary Landscape Inference and the Context-Dependence of Mutations in Beta-Lactamase TEM-1" is an example where the sequence likelihood is correlated to thermo stability and enzyme function. "Mutation effects predicted from sequence co-variation" is an application of the original Ekeberg paper to predict deleteriousness of point mutations.
metasoarous
psathyrella
matsen
lauradoepker
Kristian just explained what's going on in this paper, and it turns out to be precisely what we need to do with unseeded clusters to help @lauranoges figure out which sequences should be chosen to synthesize.
Basically, we have a big clonal family, and we want to know which sequences are likely to be actually good antibodies, and there's a bunch of literature on the structural side showing that this maximum entropy approach is an effective and reasonably motivated way to do it.
He knows one of the authors (and is working on improving the method in a few ways) so will try to get a hold of something we can run when we're ready.