ntxpred

ntxpred predicts and classifies neurotoxin proteins from amino-acid sequence information using machine-learning models to support identification and characterization of neurotoxic proteins.

Key Features:

• Machine-learning models: Uses feed-forward neural networks (FNN), recurrent neural networks (RNN), and support vector machines (SVM) for neurotoxin prediction.
• Prediction accuracies: Achieves 84.19% with FNN, 92.75% with RNN, and 97.72% with SVM on the reported dataset.
• Classification by source: SVM modules classify origin categories (eubacteria, cnidarians, molluscs, arthropods, chordates) using amino acid composition and dipeptide composition.
• Source-classification accuracies: Source classification accuracies are 78.94% using amino acid composition and 88.07% using dipeptide composition.
• Evolutionary information integration: Incorporating PSI-BLAST-derived evolutionary information with SVM increases source-classification accuracy to 92.10%.
• Classification by function: SVM models classify functional classes using amino acid and dipeptide compositions, with accuracies of 83.11% and 91.10%, respectively.
• Functional classification with PSI-BLAST: Adding PSI-BLAST-derived information raises functional classification accuracy to 95.11%.
• Training dataset: Built on 582 non-redundant, experimentally annotated neurotoxins sourced from Swiss-Prot.
• Model evaluation: All predictive and classification models were evaluated using five-fold cross-validation.

Scientific Applications:

• Toxicology studies: Facilitates identification of potential neurotoxic proteins to support toxicological assessment.
• Drug discovery: Assists in identifying candidate neurotoxic compounds from sequence data relevant to pharmacological research.
• Evolutionary biology: Supports analyses of neurotoxin evolution by integrating compositional and PSI-BLAST-derived evolutionary information.

Methodology:

Uses amino acid composition and dipeptide composition features, PSI-BLAST-derived evolutionary information, and machine-learning models (FNN, RNN, SVM) trained on 582 Swiss-Prot neurotoxins and evaluated by five-fold cross-validation.