 |
|
Date: Wednesday March the 12th, 2008
Time and location: 14:00-15:00, Building 6 (Nuffield), Room 1081 Speakers: Daniela Wieser (EBI) Abstract: Distant evolutionary relationships between proteins with low sequence similarity are difficult to recognize by current computational methods. Consequently, many sequences obtained from large-scale sequencing projects cannot be assigned to any known proteins or families despite being evolutionarily related. Most remote homology methods aim to trace subtle amino acid similarities which are exhibited by distantly related proteins. Few methods integrate three-dimensional and secondary-structures, although the latter is deemed predictable from sequence with reasonable confidence. In this talk, I speak about the benefit of using predicted secondary-structure information for common remote homology detection tasks. More specifically, I report on large-scale benchmark experiments designed to evaluate the ability of support vector machine classifiers (SVMs) to detect distant relationships between proteins. Smith-Waterman similarity scores were used with the SVMs. The scores were calculated from sequence, from observed secondary-structure and from predicted secondary-structure; the scores were used separately and in combination. The results of these experiments over all agree with the expectation that an SVM using sequence-similarity scores benefits from the inclusion of secondary-structure similarity scores. Classifiers based on observed and predicted secondary-structure similarity scores showed similar performance, indicating that secondary-structure can be confidently predicted for the proteins used in the benchmark. |
