Gelatinase Substrate Cleavage Sites Prediction Using Machine Learning and Feature Selection

• Main Authors: Hao-Chen Chang, 張浩禎
• Other Authors: 朱彥煒
• Format: Others
• Language: en_US
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/62742988869171875828

碩士 === 國立中興大學 === 基因體暨生物資訊學研究所 === 100 === There are two types in gelatinase family: Gelatinase A (MMP-2) and gelatinase B (MMP-9), which degrade extracelluar matrix. Recent studies have pointed out that the MMPs in physiological and pathological mechanisms have a variety of regulations, e.g. the immune response, tumor development and stem cell differentiation. MMP-2 and MMP-9 regulated tumor metastasis. The inhibiting drugs in clinical trials are successful in suppression of tumor metastasis; however the survival rate of patients is not improved. Scientists found the possible reasons can be 1) MMPs have high homology and similar structure, which make the specificity of inhibitors not high. 2) The substrates of MMP-2 and MMP-9 in regulatory pathways are not complete yet. The architecture of our “GelCut” prediction system has two layers. First layer builds 4 models by SVM and four types feature of binary, physical-chemical property, disorder, and solvent Accessibility and secondary structures. In particular, fold change characteristics used in this experiment. And our models compared a variety of machine learning methods to construct the second layer of prediction system. The performance of MMP-2 substrate prediction is 0.894 in MCC. MMP-9 substrate prediction is 0.644. In this study, the feature selection of physical-chemical property shown the active sites of MMP-2 and MMP-9 are different. The information is available for drug design reference. Comparing with SitePrediction, the GelCut Accuracy is 13% higher than SitePrediction. Our MMP-2 and MMP-9 substrate prediction system, GelCut, will provide biologists to information find new substrates. New possible substrates could estimate the undiscovered regulatory pathways.