Deep learning-based transcription factor binding site positioning method

Abstract

The invention provides a transcription factor binding site positioning method based on deep learning, and the method comprises the steps: carrying out one-hot coding on a DNA sequence bound with a transcription factor, obtaining a data set, and dividing the data set into a training set and a test set based on a k-fold cross validation method; constructing an FCNARRB + model based on a full convolutional network, and setting a loss function and an evaluation index; the FCNARRB + model is trained based on the training set and the loss function, the trained FCNARRB + model is used for positioning transcription factor binding sites, and the positioning result of the trained FCNARRB + model is tested and evaluated through the test set and the evaluation index. According to the method, a nucleotide-level classification model is introduced, and accurate prediction and positioning of the transcription factor binding site are realized.

Description

technical field [0001] The invention belongs to the technical field of deep learning, and in particular relates to a method for locating transcription factor binding sites based on deep learning. Background technique [0002] Proteins are ubiquitous in the human body and are a class of macromolecular compounds composed of 20 amino acids that are synthesized by cellular activities such as gene transcription and translation. Among them, there is a type of protein that specifically binds to chromosomes - DNA Binding Protein (DBP). DBP plays a key role in gene replication, recombination, chain cleavage, transcription and other processes, and is associated with cell cycle staining. A series of qualitative changes are closely related; Transcription Factors (TFs) belong to one of the DBPs, also known as trans-acting factors, which can specifically interact with the non-coding regions of the DNA sequences in the regulatory regions, and have specific effects on gene expression. Tran...

AI Technical Summary

Problems solved by technology

However, there are two shortcomings in such a model: first, the length of TFBSs in the model is fixed; second, the model assumes that the positions of TFBSs are independent of each other

[0005] In order to solve the above problems, the TFBSs recognition method based on k-mer encoding was proposed, which can encode the dependence relationship between nucleotides; but in the method based on k-mer, the gene sequence is only composed of k-mer The vector representation of the count does not consider the position of each segment in the sequence. In addition, although the position-specific sequence kernel exists, it maps the sequence to a higher-dimensional space, making it inefficient.

[0006] To sum up, for the positioning of TFBSs, the current existing method is to use the recognition algorithm to filter out possible sequences, and then use the method of probability statistics to determine the binding area. It is necessary to traverse the entire sequence to select the position with the highest probability, and the amount of data is not large. It can still cope with the time, but with the increase of the amount of data, it shows the defects of poor performance and long response time

Images