fdrMotif: identifying cis-elements by an EM algorithm coupled with false discovery rate control.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Traditional de novo motif identification methods often optimize motif models separately from statistical significance testing.
• Existing approaches may not fully account for error rates under multiple comparisons.
• Motif abundance parameters require specification or modeling in conventional methods.

Purpose of the Study:

• To introduce fdrMotif, a novel approach for de novo motif identification.
• To select maximal binding sites while controlling a user-defined false discovery rate (FDR).
• To improve upon existing motif discovery methods like MEME.

Main Methods:

• fdrMotif integrates motif model optimization (e.g., Position Weight Matrix - PWM) with significance testing at each step.
• It controls FDR by monitoring the proportion of binding sites selected across background sequences.
• An expectation-maximization (E-M)-like procedure with a novel normalization step updates the model iteratively.

Main Results:

• Simulation studies indicate fdrMotif's normalization procedure assigns higher weights to binding sites compared to other methods.
• On experimental p53 binding loci, fdrMotif identified 569 sites in 93.2% of sequences, outperforming MEME in certain aspects.
• fdrMotif demonstrated robustness to noise and higher sensitivity with similar positive predictive value compared to MEME in simulated data.

Conclusions:

• fdrMotif offers an improved method for de novo motif identification by controlling FDR.
• The approach is robust, accurate, and requires only a user-specified FDR and an initial PWM.
• fdrMotif represents a significant advancement over existing tools for identifying transcription factor binding sites.