MMGraph: a multiple motif predictor based on graph neural network and coexisting probability for ATAC-seq data.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology

Background:

• Transcription factor binding sites (TFBSs) prediction is essential for understanding transcription factor functions.
• Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) reveals open chromatin regions, offering insights into TFBSs and nucleosome positioning.
• Current tools, often based on Convolutional Neural Networks (CNNs), are limited to predicting fixed-length TFBSs from ATAC-seq data.

Purpose of the Study:

• To develop a novel motif predictor capable of identifying multiple TFBSs with varying lengths from ATAC-seq data.
• To leverage Graph Neural Networks (GNNs) as an advancement over CNNs for TFBS prediction.

Main Methods:

• Development of MMGraph, a motif predictor utilizing a three-layer GNN architecture.
• Incorporation of coexisting probability of k-mers within the GNN framework.
• Application and evaluation on 88 ATAC-seq datasets.

Main Results:

• MMGraph demonstrated superior performance compared to existing tools.
• Achieved an impressive area under the curve radar score of 2.31 across eight metrics.
• Identified 207 higher-quality multiple motifs than other available methods.

Conclusions:

• MMGraph effectively predicts multiple TFBSs of different lengths from ATAC-seq data.
• The GNN-based approach offers significant advantages over traditional CNN-based methods for motif prediction.
• MMGraph represents a valuable tool for genomic and bioinformatics research.