Protein domain detection method and system based on cost-sensitive lstm network

Abstract

The invention discloses a protein domain detection method based on a cost-sensitive LSTM network. The two-way LSTM network is adopted as a protein domain detection model; a cost-sensitive loss function is provided, and different costs are given to error division of positive and negative samples. The invention further provides a method for updating cost weight in the loss function. As for any given initial value, the cost weight can be updated in a self-adaptive mode in the network training process. The smoothing, deletion and merge operations are adopted for performing post-processing on original output of the model. The method has the advantages that the detection model adopts the two-way LSTM network, and effective modeling is performed on the long-range correlation of a protein sequence; the cost-sensitive loss function training model is used for adapting to protein domain dataset unbalanced features; the cost-weight self-adaptation updating method is introduced, and the method is applicable to any dataset; post-processing is performed on the model, and false positive results are reduced.

Description

technical field [0001] The present invention relates to a technique in the field of biomedicine, in particular to a protein domain detection method and system based on a cost-sensitive LSTM network. Background technique [0002] A protein domain is a unit in the tertiary structure of a protein and the basic unit of protein structure and function. Domains are highly conserved protein regions separated by loose border regions in proteins with multiple domains. Prediction of protein domains is an important part of protein structure research and a preparatory step for protein tertiary structure prediction. In recent years, the rapid development of data acquisition technology has produced massive biological data, and it takes a lot of cost and time to use experimental methods to segment protein domains on protein molecules. Therefore, computational methods are needed to predict protein domains to make up for the ability of biological data generation. and the gap between speed o...

AI Technical Summary

Problems solved by technology

However, this technology cannot solve the long-range correlation of protein sequences. The ability to deal with distance correlation depends on the size of the sliding window, and within the range of the sliding window, the perception between amino acid residues decays exponentially with the increase of distance; there is no Dealing with unbalanced sample sets

Images