Gene Recognition Based on Kernel Least Squares SVM

Kernel least squares support vector was used to identify genes in the small sample and nonlinear gene recognition problems. The B. subtilis whole genome sequence and related three reference data files were downloaded from GeneBank to produce a sample set including 1400 positive and 1419 negative samples. 200 positive and 200 negative samples were selected as training set and others as test set. Five features including three Z curve features, open reading frames GC ratio and length were extracted and kernel least squares support vector machine classifier was designed and optimized on training set. The results on test set showed that the recognition rate of nonlinear least squares support vector machines is up to 99.86%, which is 9.9% and 5.68% higher than linear support vector machine and fisher classifier respectively.