Vaxi-DL

Vaxi-DL predicts potential vaccine candidate antigens using pathogen-specific deep learning models to prioritize targets for preclinical vaccine development.

Key Features:

• Deep Learning Models: Four pathogen-specific deep learning models using hyper-tuned Fully Connected Layers (FCLs) predict target antigens for bacteria, protozoa, fungi, and viruses.
• Dataset Utilization: Datasets comprise antigenic and non-antigenic sequences from established vaccine candidates and the Protegen database, with biological and physicochemical properties computed using publicly available bioinformatics tools.
• Model Training and Evaluation: Datasets are scaled and normalized, split into training, validation, and testing subsets, and model performance is evaluated using accuracy, sensitivity, specificity, precision, recall, and AUC.
• Benchmarking: Performance was benchmarked against independent datasets of known target antigens and compared with VaxiJen and Vaxign-ML.
• Performance Metrics: In testing, the models identified 175 of 219 known potential vaccine candidates (PVCs) from 37 pathogens with an average sensitivity of 93%.

Scientific Applications:

• Candidate Prioritization: Prioritization of potential vaccine candidate antigens for preclinical studies across bacteria, protozoa, fungi, and viruses.
• Proteome Screening: In silico screening of proteomes to identify antigenic and non-antigenic sequences and reduce experimental workload in antigen discovery.
• Method Comparison: Comparative evaluation and benchmarking of antigen prediction outputs against VaxiJen and Vaxign-ML.

Methodology:

Training of four pathogen-specific deep learning models with hyper-tuned Fully Connected Layers on datasets of antigenic and non-antigenic sequences (from established vaccine candidates and Protegen), computation of biological and physicochemical properties using publicly available bioinformatics tools, dataset scaling/normalization and splitting into training/validation/testing subsets, evaluation via accuracy, sensitivity, specificity, precision, recall and AUC, and benchmarking against independent datasets and comparisons to VaxiJen and Vaxign-ML.