BAGIM Event: Yifan Song

Please Join Us on Thursday March 14th at 6pm at Shire on Binney Street for a talk entitled:
Got Structure?
by Yifan Song

Most computational methods that can be applied to drug discovery require a protein structure. Ideally, such structures are derived from experimental approaches such as X-ray, EM or NMR. However, frequently, an experimental structure of the specific protein of interest is not available. In such a case we turn to homology modeling.

The state-of-the-art for homology model prediction has progressed tremendously over the past two decades. Whereas useful HM models were initially only available in cases where a high-similarity homolog was available in the PDB, it is now often possible to predict such models even when the best homologs have only 15-25% similarity to the sequence whose structure is being predicted. The dramatic improvement in what is possible with HM is due to several factors, including: adoption of methods that can identify suitable low similarity templates on the basis of Markov Models, familial and other deep analysis of sequence space; the ability to incorporate multiple low-similarity templates to broadly span a target sequence; powerful approaches to model building that can handle missing structure due to lack of template, insertions and deletions; and the availability of massive compute power through processor clusters and parallelization. In many cases, the HM models that can be generated are so good—not only in terms of backbone trace, but also in such fine details as hydrogen bonding network, salt bridges, disulphide bonds, etc.—that they are suitable for downstream modeling methods.

We will present the history of HM approaches, culminating in a description of the state-of-the-art Rosetta HM workflow.