Tuesday's seminar on Statistical Physics

Cond.Matt. & Stat.Mech.Section cm at ictp.it
Fri Nov 26 12:15:56 CET 2010



JOINT ICTP/SISSA STATISTICAL PHYSICS SEMINAR

 

 Tuesday, 30 November   -   11:30 hrs.

  

Luigi Stasi Seminar Room- ICTP Leonardo Building - 1st floor

 

 

Martin WEIGT    ( Human Genetics Foundation - Torino )

 

"Inverse problems in statistical physics, and   the identification of residue-contacts in proteins"

 

Abstract

 

The structure of proteins puts constraints on the evolution of their sequence.  A random mutation of a residue in a protein (or an interacting protein pair) may lead to conflicts with the chemical and physical properties of its directly neighboring residues in the folded protein (or the assembled protein complex) and decrease the stability of the protein (or protein complex).  As a consequence, correlations in the amino-acid occupation of residues in contact emerge in the course of evolution.

Seen the huge increase in sequenced genomes, I will discuss the inverse problem: Given large multiple-sequence alignments of evolutionarily related (homologous) proteins, can we identify residue contacts by statistical analysis of amino acid frequencies? A maximum-entropy approach leads directly to the inverse Potts problem, i.e., to the inference of the coupling strengths of a disordered Potts model from a set of given equilibrium configurations.  I will formulate several  approaches to solve this problem (Boltzmann learning, mean field, TAP, susceptibility propagation), and I will show that statistical-physics based approaches lead to a huge increase in prediction power compared to more traditional correlation analysis, even if unexpected differences between the various methods emerge. I will also show that the identified contacts enable us to perform structural predictions of protein complexes, and shed light on the interaction specificity in bacterial signal transduction.


More information about the science-ts mailing list