Ab initio protein structure prediction

Project leader:  Professor LIISA HOLM,
Institute of Biotechnology, University of Helsinki

Doctoral students of the project: 
Tuomas Hätinen

Other researchers of the project: 
Swapan Mallick

Key words:  protein folding pathways, robotics, energy landscape

Project desciption and main results:
Predicting protein structure from amino acid sequence is a long-standing problem in computational biology. While scaled-up structure determination aims to solve representative structures of all major protein families within a decade, sequence classifications show a large number of singletons which cannot be linked to known families by sequence similarity. Ab initio structure prediction methods address a practical need as well as satisfying academic curiosity.

The classical formulation of the problem is to search conformational space for the minimum energy configuration. Currently, the most successful methods are based on fragment assembly, but these methods are limited by the ability to adequately sample conformation space and by the ability of the scoring function to discriminate between native and non-native conformations. Here, we propose a novel framework to remedy the sampling problem by using domain-specific knowledge to direct conformational searches to probable folding pathways. Superfolds indicate that there is a strong sequence-independent component in the shapes that a polypeptide chain may adopt. We model sequence-dependent choices in the extension paths that add the next element to a nucleus formed so far. Folding pathways will be learned based on contact patterns of secondary structure elements. Sequence patterns will be correlated with decision points in the folding pathways. Motion planning techniques will be used for loop closure. Kinetic criteria from the folding simulations will be used to select feasible folds from possible folds, as it has been suggested that negative design plays a key role in making natural sequences fast folders.

Publications: