ToxIBTL

ToxIBTL predicts toxicity of peptides and proteins using a deep learning framework that integrates the information bottleneck principle, transfer learning, evolutionary information, and physicochemical properties to support the development of safer peptide therapeutics.

Key Features:

• Deep Learning Framework: Implements a deep learning model that integrates the information bottleneck principle and transfer learning for toxicity prediction.
• Information Bottleneck Principle: Optimizes feature representations by retaining relevant information while reducing redundancy for toxicity prediction.
• Transfer Learning: Transfers knowledge from protein data to improve peptide feature representation and prediction performance.
• Evolutionary Information and Physicochemical Properties: Incorporates evolutionary information and physicochemical properties of peptide sequences into the feature representation scheme.
• Benchmark Performance: Demonstrated to outperform existing state-of-the-art methods for peptide toxicity prediction and to show competitive performance on protein datasets.

Scientific Applications:

• Peptide Drug Development: Supports the design of safer peptide-based therapeutics by predicting peptide toxicity.
• Early-stage Toxicity Screening: Aids identification and modification of potentially harmful peptides early in the development process.
• Advancing Therapeutics: Contributes to advancing peptide drugs from research toward clinical applications by informing toxicity-related decisions.

Methodology:

ToxIBTL trains a deep learning model on datasets comprising both peptides and proteins, integrates evolutionary information and physicochemical properties into feature representations, applies the information bottleneck principle to optimize features, and uses transfer learning from protein data to improve peptide predictions.