A sequence-based two-level method for the prediction of type I secreted RTX proteins

Area of Science:

• Microbiology
• Molecular Biology
• Bioinformatics

Background:

• Gram-negative bacteria utilize the type I secretion system (T1SS) for extracellular protein translocation.
• Type I secreted RTX proteins (T1SRPs) are crucial for pathogen-host interactions, but experimental identification is challenging.
• High-throughput genomics necessitates computational methods for predicting T1SRPs from amino acid sequences.

Purpose of the Study:

• To develop and validate a computational method for identifying T1SRPs using sequence-derived features.
• To investigate sequence features predictive of T1SRPs at both full-length and C-terminal secretion signal levels.

Main Methods:

• A two-level random forest (RF) algorithm was employed.
• Sequence-derived features, including RTX repeats and C-terminal amino acid/secondary structure compositions, were utilized.
• Five-fold cross-validation and benchmarking on an independent dataset were performed for evaluation.

Main Results:

• The presence of RTX repeats effectively distinguished T1SRPs at the full-length sequence level.
• A C-terminal sequence segment (last 20-30 amino acids) showed predictive value for T1SRP secretion.
• High accuracies of 97% (full-length) and 89% (secretion signal) were achieved, with 63/74 and 66/74 T1SRPs correctly predicted on an independent dataset.

Conclusions:

• The developed RF-based method accurately predicts T1SRPs using sequence information.
• This computational approach aids in elucidating T1SS mechanisms and facilitates experimental design.
• The findings are valuable for understanding bacterial pathogenesis and protein secretion.