SortPred

SortPred predicts bacterial sortase enzymes and classifies them into classes A-F using sequence-derived feature descriptors and machine learning for sequence annotation.

Key Features:

• Two-Layer Prediction Model: A two-layer approach first discriminates sortase versus non-sortase sequences and then assigns predicted sortases to one of six classes (A-F).
• Random Forest Classifier: Random forest classifiers were trained for each descriptor and used as the primary prediction algorithm.
• Feature Descriptors and Encoding Algorithms: Development involved investigation of 31 feature descriptors across five encoding algorithms to define model inputs.
• Benchmarking Dataset: An original benchmarking dataset was constructed to train and validate models.
• Model Evaluation and Selection: Models were evaluated by cross-validation and an independent dataset, and final models were selected based on consistent performance.

Scientific Applications:

• Identification of Bacterial Sortases: Facilitates identification and annotation of bacterial sortase enzymes from protein sequences, including those from Gram-positive bacteria.
• Class Assignment and Functional Analysis: Enables assignment to classes A-F to support downstream functional, comparative, and evolutionary analyses.
• Sortase Inhibitor Development: Supports selection of candidate sortases relevant to the development of sortase inhibitors and investigations of substrates with pathological implications.

Methodology:

An original benchmarking dataset was constructed; 31 feature descriptors were analyzed using five encoding algorithms; each descriptor was used to train random forest classifiers; models were evaluated by cross-validation and an independent dataset and final models were selected based on consistent performance.