MODER2

MODER2 discovers monomeric and dimeric transcription factor (TF)-binding motifs using first-order Markov models (adjacent dinucleotide matrices, ADMs) to model dependencies between adjacent nucleotides.

Key Features:

• First-order Markov models (ADMs): Models dependencies between adjacent nucleotides using adjacent dinucleotide matrices to capture dinucleotide context that position-specific probability matrices (PPMs) cannot.
• Monomeric and dimeric motif estimation: Simultaneously estimates models for monomeric and dimeric TF-binding motifs.
• ADM-based mixture model: Incorporates monomers and corresponding dimers within an ADM-based mixture framework.
• Dimeric structure modeling: Explicitly represents dimeric structure including spacing and orientation to capture cooperative dimerization effects.
• Expectation maximization learning: Fits model parameters using an expectation maximization algorithm from training data and seeds.
• Validation and benchmarking: Validated on HT-SELEX experiments and generated datasets and benchmarked against PPMs across 314 tested transcription factors or DNA-binding domains from bHLH, bZIP, ETS, and Homeodomain families.
• Implementation: Implemented in C++ with a dependency on the Boost library and tested on Linux platforms.

Scientific Applications:

• TF-binding motif discovery: Identification of monomeric and dimeric transcription factor-binding motifs from sequence data.
• Cooperative dimerization analysis: Investigation of cooperative binding effects by modeling spacing and orientation between monomers in dimers.
• High-throughput data analysis: Interpretation and modeling of HT-SELEX and generated motif datasets.
• Motif model benchmarking: Comparative evaluation of ADM mixture models versus PPMs across multiple TF families (bHLH, bZIP, ETS, Homeodomain).

Methodology:

Uses first-order Markov models (ADMs) within an ADM-based mixture model of monomers and dimers and fits parameters by expectation maximization from training data and seeds; validation performed on HT-SELEX and generated datasets and comparisons made to PPM models across 314 TFs from bHLH, bZIP, ETS, and Homeodomain families.