<p>A central problem in computational biology is the classification of related proteins into functional and structural classes based on their amino acid sequences. Several methods exist to detect related sequences when the level of sequence similarity is high, but for very low levels of sequence similarity the problem remains an unsolved challenge. Most recent methods use a discriminative approach and train support vector machines to distinguish related sequences from unrelated sequences. One successful approach is to base a kernel function for a support vector machine on shared occurrences of discrete sequence motifs. Still, many protein sequences fail to be classified correctly for a lack of a suitable set of motifs for these sequences. We introduce a motif kernel based on discrete sequence motifs where the motifs are synthesised using genetic programming. The motifs are evolved to discriminate between different families of evolutionary origin. The motif matches in the sequence data sets are then used to compute kernels for support vector machine classifiers that are trained to discriminate between related and unrelated sequences. When tested on two updated benchmarks, the method yields significantly better results compared to several other proven methods of remote homology detection. The superiority of the kernel is especially visible on the problem of classifying sequences to the correct fold. A rich set of motifs made specifically for each SCOP superfamily makes it possible to classify more sequences correctly than with previous motif-based methods.</p>
| Identifer | oai:union.ndltd.org:UPSALLA/oai:DiVA.org:ntnu-9481 |
| Date | January 2006 |
| Creators | Håndstad, Tony |
| Publisher | Norwegian University of Science and Technology, Department of Computer and Information Science, Institutt for datateknikk og informasjonsvitenskap |
| Source Sets | DiVA Archive at Upsalla University |
| Language | English |
| Detected Language | English |
| Type | Student thesis, text |
Page generated in 0.0123 seconds