GNNMF: a multi-view graph neural network for ATAC-seq motif finding.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) identifies open chromatin and multiple transcription factor binding sites (TFBSs).
• Deep learning, particularly Convolutional Neural Networks (CNNs), has been used to find motifs in ATAC-seq data, but is limited to fixed motif lengths.
• Graph Neural Networks (GNNs) offer potential for finding ATAC-seq motifs of varying lengths, but existing models overlook sequence relationships and require parameter optimization.

Purpose of the Study:

• To develop a novel GNN model, GNNMF, for identifying multiple ATAC-seq motifs of varying lengths.
• To improve the accuracy and scope of TFBS prediction using ATAC-seq data.
• To address limitations of existing DL models in motif discovery from ATAC-seq data.

Main Methods:

• Developed GNNMF, a novel GNN model integrating GNNs with background coexisting probability.
• Constructed a multi-view heterogeneous graph using ATAC-seq sequences.
• Utilized background coexisting probability and iterloss for motif discovery and parameter optimization.

Main Results:

• GNNMF demonstrated improved performance on human and mouse datasets, with radar scores increasing by 4.92% and 6.81% respectively across eight metrics.
• The model successfully identified a greater number of ATAC-seq motifs compared to existing methods.
• GNNMF achieved superior performance in TFBS prediction and ATAC-seq motif finding.

Conclusions:

• GNNMF is a novel and effective model for finding multiple ATAC-seq motifs of different lengths.
• The developed model enhances TFBS prediction accuracy and motif discovery capabilities.
• GNNMF represents a significant improvement over existing methods for ATAC-seq motif analysis.