Membrane Protein Identification via Multiple Kernel Fuzzy SVM

Abstract

Membrane proteins are an important part of daily life activities in biological information. Predicting membrane proteins can improve drug targeting accuracy and artificial intelligence-assisted drug progression. Traditional methods such as X-ray and MRI are more accurate but consume huge human and material resources, and as science progresses, traditional experimental methods have become more and more difficult to match the needs of experts. In this paper, from the perspective of machine learning, pseudo-PSSM (PsePSSM), averaging block (AvBlock), discrete cosine transform (DCT), discrete wavelet transform (DWT) and histogram of oriented gradients (HOG) are used, and then features are extracted via position scoring matrix (PSSM). An evolutionary feature and fuzzy support vector machine based membrane protein prediction model is proposed. The results show that this method has better prediction and higher accuracy than other methods on two benchmark data sets, TRAIN1 and TRAIN2, reaching 90.6% and 89.7% accuracy, respectively.