Predicting protein structure by solving the inverse Potts problem: a pseudo-likelihood approach
by
Magnus Ekeberg(KTH)
→
Europe/Stockholm
122:026
122:026
Description
When analyzing systems of interacting elements, disentangling the direct interactions from the ones who are network-mediated is an intrinsically complex task. This issue has been proven critical in Protein Structure Prediction, an area where data from many sequenced (evolutionarily linked) proteins can be used to infer information about the three-dimensional structures of the proteins in a family. The disentangling itself requires one to solve what is known as the inverse Potts problem, a task for which naive mean-field theory has previously been used. More sophisticated methods than naive mean-field exist, and we discuss one such known as pseudo-likelihood maximization, which seems to provide a better alternative to perform this task.
