SEME

SEME (Sampling with Expectation Maximization for Motif Elicitation) discovers de novo sequence motifs, particularly transcription factor (TF) binding sites, by integrating a probabilistic mixture model, importance sampling, and Expectation-Maximization to learn motif sequences, position preferences, and sequence-rank preferences.

Key Features:

• Probabilistic Mixture Model: Employs a pure probabilistic mixture model to represent motif and background binding characteristics.
• Expectation Maximization (EM) Algorithm: Uses EM to simultaneously learn sequence motifs, motif positions, and sequence-rank preferences from input sequences.
• Variable Motif Length Extension: Adapts motif length during analysis to detect motifs of varying sizes.
• Importance Sampling: Incorporates importance sampling to focus computation on high-probability regions of the sequence space.
• Position and Sequence-Rank Preferences: Models position preference and sequence-rank preference for motifs, including those of co-regulatory TFs (coTFs).

Scientific Applications:

• De novo TF binding-site discovery: Demonstrated improved motif discovery performance on 75 large-scale synthetic datasets, 32 metazoan compendium benchmark datasets, and 164 ChIP-Seq libraries.
• Co-regulated TF (coTF) motif identification: Identified a significantly higher number of correct coTF motifs across 15 ChIP-Seq libraries with predictions closely matching known motifs.
• Experimental validation: Learned position and sequence-rank preferences and motif predictions have been validated through additional ChIP-Seq experiments.

Methodology:

Applies a pure probabilistic mixture model with importance sampling and an Expectation-Maximization procedure that simultaneously learns sequence motifs, motif position preferences, and sequence-rank preferences, and incorporates a variable motif length extension for each coTF.