Discriminative motif optimization based on perceptron training.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Accurate transcription factor (TF) binding site motif generation from next-generation sequencing data, like ChIP-seq, is challenging due to large datasets and long sequences.
• Traditional motif finders struggle with speed and may not fully utilize input sequence information, leading to suboptimal motifs.

Purpose of the Study:

• To develop a tool that rapidly improves the accuracy of TF binding site motifs.
• To enhance existing motifs using a discriminative strategy that maximizes motif discriminating power.

Main Methods:

• Introduced discriminative motif optimizer (DiMO), a tool that refines seed motifs using positive and negative sequence databases.
• Employed the area under the receiver-operating characteristic curve (AUC) as a performance metric.
• Utilized a perceptron training strategy for rapid AUC maximization.

Main Results:

• DiMO significantly improved motifs identified by nine different motif finders across 87 TFs from human, Drosophila, and yeast.
• AUC was improved for approximately 90% of TFs on test sets.
• The magnitude of AUC increase reached up to 39%.

Conclusions:

• DiMO offers a rapid and effective method for enhancing TF binding site motif accuracy.
• The discriminative approach maximizes the utility of sequence data for improved motif discovery.