MILLIPEDE

MILLIPEDE identifies transcription factor binding sites by integrating DNase digestion data with transcription factor binding specificity information to survey genomic locations of multiple TFs within a single experiment.

Key Features:

• Data integration: Integrates DNase digestion signals with transcription factor binding specificity information to evaluate genomic locations of multiple TFs.
• Performance superiority: Outperforms CENTIPEDE, marginally in human datasets and dramatically in yeast, increasing the average auROC across 20 TFs from 74% to 94%.
• Logistic regression framework: Uses a logistic regression statistical model as the core supervised learning component.
• Supervision versatility: Provides supervised, partially supervised, and completely unsupervised variants with performance close to supervised versions.
• Parameter efficiency: Requires at least an order of magnitude fewer parameters than CENTIPEDE.

Scientific Applications:

• Transcription factor binding site identification: Detects TF binding sites across multiple transcription factors using DNase digestion and specificity data.
• Gene regulation studies in human and yeast: Enables comparative analyses of transcriptional regulation mechanisms, with notably improved performance in yeast.
• High-throughput genomic analyses: Applicable to large-scale surveys of TF binding using DNase digestion experiments.

Methodology:

MILLIPEDE applies a logistic regression model that integrates DNase digestion signals with transcription factor binding specificity and is available in supervised, partially supervised, and unsupervised variants.