CSM-Toxin

CSM-Toxin predicts potential toxicity of peptides and proteins from their primary amino acid sequences to support early identification of toxic biologics.

Key Features:

• Primary Sequence Reliance: Predicts toxicity using only the protein or peptide primary amino acid sequence without requiring structural or functional annotations.
• Deep Learning (NLP-inspired): Employs a deep learning model inspired by natural language processing that treats amino acid residues as words and protein sequences as sentences.
• Curated Experimental Dataset: Trained on a manually curated dataset derived from high-quality experimental data on peptide and protein toxicities.
• Validation and Performance: Validated via cross-validation and multiple non-redundant blind tests, achieving a Matthews Correlation Coefficient (MCC) of up to 0.66.

Scientific Applications:

• Early toxicity screening: Identify peptides and proteins with potential toxic properties early in biologic development workflows.
• Therapeutic candidate prioritization: Prioritize peptide- and protein-based candidates for experimental follow-up based on predicted toxicity risk.
• Clinical trial risk reduction: Flag candidates with potential toxicity to reduce the risk of clinical-trial failures due to unforeseen toxicities.

Methodology:

Uses primary amino acid sequences as input to a deep learning model inspired by NLP, trained on a manually curated experimental toxicity dataset and evaluated by cross-validation and multiple non-redundant blind tests reporting MCC up to 0.66.