SeqGL Identifies Context-Dependent Binding Signals in Genome-Wide Regulatory Element Maps.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genome-wide assays like ChIP-seq, DNase-seq, and ATAC-seq generate data that implicitly contains DNA sequence signals for multiple transcription factors (TFs).
• Discovering these TF binding motifs is crucial for understanding gene regulation.

Purpose of the Study:

• To introduce SeqGL, a novel de novo motif discovery algorithm designed to identify multiple TF sequence signals from various genomic profiles.
• To evaluate SeqGL's performance against traditional motif discovery tools.

Main Methods:

• SeqGL employs a k-mer feature representation and group lasso regularization to train a discriminative model.
• The model extracts sequence signals that differentiate TF-bound or open chromatin regions from flanking sequences.
• The algorithm was benchmarked on over 100 ChIP-seq experiments and tested on DNase-seq and ATAC-seq data.

Main Results:

• SeqGL demonstrated superior discriminative accuracy compared to traditional motif discovery tools in benchmark tests.
• The algorithm successfully identified ChIP-seq validated sequence signals missed by other methods, indicating higher sensitivity.
• SeqGL scales effectively for large datasets like DNase-seq and ATAC-seq, identifying multiple sequence signals.

Conclusions:

• SeqGL offers improved discriminative accuracy and sensitivity for detecting DNA sequence signals underlying regulatory elements.
• The algorithm can be integrated with multitask learning to explore genomic and cell-type specific TF binding determinants.
• SeqGL represents a significant advancement in motif discovery for analyzing diverse genomic datasets.