ToxinPred2

ToxinPred2 predicts protein toxicity using sequence-based analyses to identify toxic and non-toxic proteins for therapeutic development.

Key Features:

• Training and Evaluation: Trained, tested, and evaluated on three datasets derived from the latest release of SwissProt with internal validation on 80% and external validation on 20% of the data.
• BLAST-based Similarity: Uses Basic Local Alignment Search Tool (BLAST)-based sequence similarity to identify proteins related to known toxins.
• Motif-EmeRging and with Classes-Identification-based Motif Search: Detects motifs associated with toxicity using the Motif-EmeRging and with Classes-Identification-based motif search.
• Machine Learning Models: Employs machine learning models tuned to balance sensitivity and specificity for toxicity prediction.
• Hybrid Method: Integrates BLAST, MERCI motif search, and machine learning into a hybrid method achieving an area under the ROC curve of approximately 0.99 and a Matthews correlation coefficient of 0.91 on validation datasets.
• General Applicability: Predicts toxicity across proteins from diverse sources for broad research applicability.

Scientific Applications:

• Therapeutic protein and peptide development: Enables early identification of potentially toxic protein and peptide candidates during therapeutic development.
• Preclinical safety assessment: Supports selection and prioritization of protein candidates by predicting potential toxicity to reduce adverse selections.

Methodology:

Combines BLAST-based similarity searching, Motif-EmeRging and with Classes-Identification-based motif search, and machine learning models into a hybrid predictive approach.