DISCOVER

DISCOVER predicts transcription factor binding sites in metazoan genomes using a discriminative conditional random field (CRF) model for supervised motif discovery.

Key Features:

• Conditional Random Fields Framework: Uses a discriminative CRF model to optimize the predictive probability of motif presence within large sequences.
• Joint biological feature modeling: Considers the joint effect of multiple biological features to improve motif prediction accuracy.
• Integration of genetic and epigenetic factors: Incorporates clade-specific evolutionary parameters and proximity to coding regions as explicit input features.
• CRM architecture modeling: Models the grammatical organization and architectural context of motifs within cis-regulatory modules (CRMs).
• False positive reduction: Employs a discriminative approach to reduce false positives arising from motif degeneracy and spurious nucleotide signals.

Scientific Applications:

• TFBS prediction: Predicts transcription factor binding sites in metazoan genomes and was evaluated on simulated CRMs and real Drosophila sequences.
• Regulatory signal discovery: Enables decoding of complex regulatory signals and cis-regulatory architecture in metazoan genomes.
• Performance benchmarking: Demonstrated a reported 22% improvement in F1 score over prior models on the stated evaluations.

Methodology:

Discriminative conditional random field (CRF)-based supervised motif discovery that integrates clade-specific evolutionary parameters and coding-proximity features, models CRM grammatical organization, and was evaluated on simulated CRMs and real Drosophila sequences.