BLOSUM Coding Schemes of Prediction of Protein Stability upon Point Mutations Based on Amino Acid Sequences
基于进化信息改进蛋白质定点突变稳定性预测准确率

Different approaches were developed to predict the stability change of protein point mutation. Some of them used the sequence information while others used the structure information of the protein. The authors touched the problem with the protein sequence information. Encoding schemes of the machine learning approaches towards the problem differ from one to another. Sparse encoding and amino acid property encoding prevailed in the field. Physicochemical properties such as hydropathy, flexibility, electronic charge concentration and the isotropic surface area (ISA) of amino acids were picked up as input attributes. Evolution information which is embedded in BLOSUM matrices were tested in the paper. An improvement over previous encoding schemes was found in our experiments. Machine learning techniques such as SVM (Support vector machines) and artificial neural networks were used for evaluating the new encoding schemes. Results show that the BLOSUM62 encoding scheme outperform the sparse encoding scheme and amino acid property encoding scheme.