Tuesday, October 6, 2015
1:00 pm, MRB 100C
Dr. Arne Elofsson
Science for Life Laboratory, Stockholm University, Solna, Sweden
Improved protein structure predictions using PconsC
The ultimate goal of structural bioinformatics research is to provide a complete structural map of all macromolecules and their interactions within a cell. Knowledge about the structure of proteins and other macromolecules is essential for our understanding of biological processes. Proteins do most of the work in a cell and therefore the studies of proteins structure has been an essential part of life science research during the last decades. The rapid growth of determined protein structures has made it possible to build homology models for many proteins.
However, surprisingly the exponential reduction in sequencing costs has also been fundamental for the progress since it allows more distant homologies to be detected and nowadays be used to predict contacts in proteins reliably. Here, I will describe our work on the development of our contact predictor, PconsC. By using a combination of direct coupling analysis, classical machine learning and deep learning approaches we are now able to accurately predict the contacts of protein families with as little as 100 effective sequences.
Futher, most proteins do not act alone but through interactions with other proteins (and other molecules). Therefore, it is essential to understand not only the structure of a protein but also its interactions. Here, systems biology approaches are often used to understand what interactions are made but these studies mostly ignore the atomistic details about the interactions, i.e. how, the interactions are made. Also here the accurate prediction of inter-residue contacts can provide valuable information. I will discuss the outlook for how contact prediction can be used here.