DIpartite: A tool for detecting bipartite motifs by considering base interdependencies

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Transcription factor binding sites (TFBSs) are crucial for gene regulation.
• Bipartite motifs, characterized by two blocks separated by variable gaps, represent a significant class of TFBSs.
• Traditional Position Weight Matrix (PWM) models do not fully capture base interdependencies.

Purpose of the Study:

• To develop a novel computational tool, DIpartite, for *ab initio* motif detection.
• To enhance TFBS prediction by incorporating dinucleotide weight matrices (DWMs) to account for base interdependencies.
• To specifically improve the detection of bipartite motifs with variable gap lengths.

Main Methods:

• Developed DIpartite, a tool employing Gibbs sampling and Shannon's entropy minimization.
• Incorporated dinucleotide weight matrices (DWMs) alongside traditional PWMs for motif representation.
• Utilized test datasets including CRP in E. coli, sigma factors in B. subtilis, and human promoter sequences for evaluation.

Main Results:

• DIpartite accurately predicts bipartite motifs by considering dinucleotide interdependencies.
• Performance evaluation showed DIpartite is equivalent or superior to existing tools (MEME, BioProspector, BiPad, AMD), particularly for motifs with variable gaps.
• The tool allows users to specify motif lengths, gap lengths, and the type of weight matrix (PWM or DWM).

Conclusions:

• DIpartite offers an effective *ab initio* method for TFBS detection, especially for bipartite motifs.
• The incorporation of DWMs provides a more nuanced representation of sequence motifs.
• DIpartite is available as an open-source tool, facilitating further research in motif discovery.