Prediction of Anti-parallel and Parallel Beta-sheets using Conditional Random Fields

In this paper, we present a new algorithm for parallel and anti-parallel beta-sheet prediction using conditional random fields. In recent years, various approaches have been proposed to capture the long-range interactions of beta-sheets.  However, most of them are not very successful: either the learning models are not general enough to capture the non-local information, or the features they used do not contain the information, for example, the window based profiles. Our new method has the advantages over previous methods in two aspects: (1) It takes into account both the local information and long-range interaction information (2) The condition random fields are powerful models that are able to capture long-range interaction features.  The experimental results show that our algorithm performs significantly better than the state-of-art secondary structure prediction methods.