Motif mining method based on sequence and shape information deep fusion

Abstract

The invention relates to a motif mining method based on sequence and shape information deep fusion. The method comprises the following steps of: S1, constructing a deep embedded convolutional neural network model, and carrying out space alignment mixing on a DNA sequence and shape information to serve as input of the neural network model; S2, training the deep embedded convolutional neural network model to obtain a motif binding strength predicted value; and S3, evaluating the performance of the constructed deep embedded convolutional neural network model based on a regression coefficient R2 between the predicted value and actual binding strength. According to the method, the advantages of the convolutional neural network in the aspect of feature extraction are utilized, and deep fusion of sequence and shape features is realized.

Description

technical field [0001] The present invention relates to the technical field of computer recognition and deep learning, in particular to a motif mining method based on deep fusion of sequence and shape information. Background technique [0002] The binding behavior of transcription factors plays an important role in the regulation of gene expression, and the identification of transcription factor binding sites is of great significance for understanding the binding mechanism and related cellular activities. However, transcription factor binding is a delicate biophysical process with many influencing factors and difficult modeling. To this end, researchers have developed a variety of binding site prediction models. The position weight matrix models the sequence-specific preference of transcription factor binding through a probabilistic statistical model, and each column element value in the matrix represents the probability distribution of the four nucleotides {A, C, G, and T}...

AI Technical Summary

Problems solved by technology

However, such traditional methods often ignore the sequence relationship between K-mers, and machine learning models are limited by computing performance and optimization methods, and cannot make full use of massive sequencing data.

Images