Effect of positional dependence and alignment strategy on modeling transcription factor binding sites

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Existing methods for transcription factor binding site (TFBS) recognition struggle with variable binding site lengths and often exclude known sites.
• The influence of Information Content (IC) and nucleotide positional dependence on modeling variable-length TFBS remains underexplored.

Purpose of the Study:

• To introduce ML-Consensus (Mixed-Length Consensus), a novel consensus model for variable-length TFBS prediction.
• To evaluate the impact of incorporating IC and Pairwise Score (PS) on TFBS prediction accuracy.

Main Methods:

• Developed ML-Consensus, a model that accommodates all known binding sites, regardless of length.
• Utilized Pairwise Score (PS) to quantify nucleotide positional dependence within TFBS alignments.
• Performed cross-validation on six species' datasets from TRANSFAC, analyzing results with ROC curves and Wilcoxon tests.

Main Results:

• Incorporating IC and PS into ML-Consensus significantly enhanced prediction accuracy.
• Demonstrated the presence of a core binding region characterized by nucleotide co-occurrence within the core length.

Conclusions:

• The findings suggest that ML-Consensus effectively models variable-length TFBS.
• Further research into efficient multiple sequence alignment algorithms for TFBS is warranted to optimize prediction.