DISCOVER: a feature-based discriminative method for motif search in complex genomes

Area of Science:

• Computational genomics
• Bioinformatics
• Systems biology

Background:

• Identifying transcription factor binding sites (TFBSs) in metazoan genomes is challenging due to motif degeneracy and complex cis-regulatory module (CRM) organization.
• Existing pattern matching methods yield high false positive rates on large genomic datasets and noisy position weight matrices.

Purpose of the Study:

• To develop a novel computational method for accurate TFBS prediction in metazoan genomes.
• To improve upon existing methods by utilizing both CRM architecture and individual motif features.

Main Methods:

• A discriminative probabilistic model using conditional random fields (CRFs) was employed.
• The model integrates biologically meaningful features like clade-specific evolutionary parameters and proximity to coding regions.
• It optimizes the predictive probability of motif presence based on the joint effect of these features.

Main Results:

• The proposed model overcomes limitations of earlier statistical formalisms sensitive to spurious signals.
• Evaluated on simulated and real Drosophila sequences, the method significantly outperforms state-of-the-art models.
• Achieved a 22% improvement in F1 score compared to existing TFBS prediction methods.

Conclusions:

• The developed CRF-based method offers a more robust and accurate approach for TFBS identification in metazoan genomes.
• This advancement aids in understanding complex gene regulation by reducing false positives in genomic analyses.