DeepTFactor

DeepTFactor predicts whether a protein functions as a transcription factor to identify sequence-specific DNA-binding proteins that regulate gene expression.

Key Features:

• Deep learning model: Utilizes a convolutional neural network to extract and analyze protein features.
• Taxonomic scope: Identifies both eukaryotic and prokaryotic transcription factors.
• Performance: Reports F1 scores of 0.8154 for eukaryotes and 0.8000 for prokaryotes.
• Interpretability: Detects DNA-binding domains and latent features supported by analysis of gradients of prediction scores relative to input data.
• Experimental validation: Predicted candidate TFs in Escherichia coli K-12 MG1655, including 84 y-ome candidates, and characterized genome-wide binding sites of YqhC, YiaU, and YahB.
• Large-scale predictions: Generated predictions of 4,674,808 transcription factors from 73,873,012 protein sequences across 48,346 genomes.

Scientific Applications:

• Discovery of novel TFs: Enables identification of transcription factors lacking sequence homology to known DNA-binding domains.
• Genome annotation: Prioritizes candidate transcription factors within genomes such as Escherichia coli K-12 MG1655, including y-ome genes.
• Regulatory network characterization: Supports mapping and experimental characterization of genome-wide TF binding sites.
• Comparative genomics: Provides large-scale TF predictions across thousands of genomes to study organismal regulatory systems.

Methodology:

Uses a convolutional neural network to extract and analyze protein features and applies gradient analysis of prediction scores relative to input data for interpreting detected features.