Computational Biology and Bioinformatics Discussion Group (1)

Organizer: Andras Fiser, Albert Einstein College of Medicine

Speakers: Ao Ma, Albert Einstein College of Medicine; Donald Petrey, Columbia University; Andrey Rzhetsky, Columbia University

The Bioinformatics and Computational Biology Discussion Group brings together diverse institutions and communities to share new and relevant information at the frontiers of the interrelated fields of bioinformatics and computational biology. Recent topics have included "Benchmarking and Improving the Accuracy of Comparative Modeling of Protein Structures," "Integrated Statistical Modeling of Gene Expression Data," and "Estimating SNP Haplotype Frequencies from DNA Pools."

Abstracts

Systematic methods for understanding reaction dynamics in complex biophysical systems
Ao Ma
Albert Einstein College of Medicine

Conformational dynamics plays essential role in the function of complex biomolecular systems. Consequently, it is very important to understand the detailed molecular mechanism underlying these processes. At the same time, achieving such an understanding also poses significant challenges since we need to obtain realistic reactive trajectories and to extract the essential dynamic information contained in these trajectories.

Here I will present a novel method for generating an unbiased trajectory from a biased one by continuously decreasing the bias in subsequent simulations. I will also present an automatic method for identifying reaction coordinates from the transition trajectories.

Representing and exploiting protein structural similarity
Donald Petrey
Columbia University

The ability to derive useful biological information about an experimentally uncharacterized protein, either based on a computational model of the sequence or with knowledge the protein structure from a structural genomics (SG) initiative, relies heavily on what one can say about proteins that are structurally similar to the model or SG result. Of course, this in turn depends on how one quantifies and represents structural similarity. Recent work in our group has demonstrated that existing structure classification schemes such as SCOP and CATH, with their reliance on poorly defined concepts such as "fold" and “domain”, can obscure genuinely useful functional an evolutionary relationships. Indeed, the concept of fold is being phased out of the newest versions of these databases. I will discuss these issues and describe a recently developed alternative representation of structural similarity. I will also discuss the use this representation can be used in protein structure and function prediction.