T4SEfinder

T4SEfinder predicts bacterial type IV secreted effectors (T4SEs) from protein sequences to enable genome-scale identification and re-annotation of effectors associated with type IV secretion systems (T4SSs) and bacterial virulence.

Key Features:

• Sequence embedding (TAPE): Employs embeddings derived from the pre-trained protein language model TAPE (Tasks Assessing Protein Embeddings) for feature representation.
• Training dataset (SecReT4): Trained on the SecReT4 database updated with newly experimentally verified T4SEs.
• Model architecture (MLP): Implements a multi-layer perceptron (MLP) classifier optimized through comprehensive comparisons against several candidate models.
• Performance optimization: Model selection and tuning based on performance comparisons, resulting in prediction accuracy that slightly surpasses existing tools.
• Computational efficiency: Capable of classifying approximately 5,000 protein sequences in about three minutes, enabling whole-genome-scale analysis.
• Genome-scale detection: Supports detection and re-annotation of T4SEs and T4SS-associated effectors across sequenced bacterial genomes.

Scientific Applications:

• Genome-scale T4SE identification: Detection of T4SEs in pathogenic bacterial proteomes to map effector repertoires.
• Re-annotation of secretion systems: Re-annotation of type IV secretion systems and associated effector proteins across sequenced bacterial genomes.
• Virulence factor discovery: Identification of candidate virulence factors to support studies of bacterial pathogenesis and host–pathogen interactions.

Methodology:

Uses sequence embeddings from the pre-trained protein language model TAPE, trains a classifier on SecReT4 entries updated with experimentally verified T4SEs, and implements an optimized multi-layer perceptron (MLP) selected via comprehensive performance comparisons.