Prediction Of Splice Sites Based On SVM

As more and more genome data is generated, it's the main target to make use ofbioinformatic methods to study the function and the expression process of gene. Andsplice sites of eukaryotic cells is an important factor in the expression of gene. So it isa very important part of gene prediction and helpful to understand gene function.We can consider it as a classification problem, which is studied by making use ofthe features near splice sites to distinguish the real sites from DNA sequences. At first,we make use of fuzzy support vector machine based on mixture kernels to recognizethe splice sites, and compare the results with basic support vector machine. Itsrecognition rate is higher the basic one kernel SVM methods. Then the multi-SVMs isput forward to resolve this classification problem. This method synthesizes differentprediction information with different weights to get a last result and optimize theseparameters by Simulated Annealing algorithm. We found this simple method can getbetter results than the basic support vector machine.