| Protein structure prediction is an important part of research work about the protein structure and function,which is of great significance on protein drug design,pharmaceutical and other aspects.If the structure of some proteins in the family of homologous is known,it is possible to predict the structure of other proteins that are unknown.By sequence alignment,unequal sequences length can be changed into equal-length sequences by inserting vacancies,which represent the evolution of the sequence from the same ancestor through insertion and deletion,and so on,and reflects the variation in biological evolution,mutation.The presence of vacancies can have a significant impact on the scale and accuracy of homologous protein modeling.So it is very important to study the missing value in protein sequence alignment.The filling of protein missing data has been achieved by some methods,the nearest neighbor algorithm,and the self-organizing neural network algorithm.Both methods provided a reasonable filling for the protein missing data,and increased from 62.9% to 82.7% on the average exploration scale.The accuracy of the study was improved from 1.65 ? to 0.88 ?.However,due to the complexity of protein structure space,the computation space of protein sampling is very large,which makes the calculation process more time-consuming.Therefore,we hope to improve the speed of calculation and reduce the computation time under the premise that protein missing value can be filled reasonably.Based on the nonlinear principal component algorithm of auto-associative neural networks(AANN),and considering the complexity of the spatial structure of protein sequence sampling and the growth rate of the protein column database,we use a modified inverse nonlinear network model(Inverse NLPCA Model)to achieve the missing value of filling and efficiency,and the network model using conjugate gradient algorithm optimization to further accelerate the calculation efficiency. |
PDF Full Text Request