Department of Computer Science and Engineering Taiyuan Normal University, Taiyuan, 030012, China

---

Abstract—The scarce but consistent chance of getting false positive matches [1], [2] in protein database search [3] has always casted a shadow over the reliability of results. The situation can be helped by viewing the protein data from a descriptive and the probabilistic framework, together. Using the conventional approach as the first stage, top down protein data is descriptively searched for proteins and the results are scored and ranked, using a top down protein search engine. We then suggest applying Support Vector Machine, (SVM) as a second stage probabilistic scoring system, to the first stage protein database search results so as to further enhance protein classification. For SVM scoring, features are extracted from the top down data and a feature table is constructed. An SVM using Radial Basis Function is trained with this feature table. Later classification is performed on the test data using this SVM. The classification can then be viewed together with the previously calculated search engine score and a reordering of top ranked proteins may be done.

Index Terms—False positive, protein, SVM.

[PDF]

Cite: Huijuan Guo, "False Positives Reduction in Top-down Protein Informatics using Support Vector Machines," Journal of Computers vol. 7, no. 4, pp. 900-912, 2012.