DeepImmuno

DeepImmuno predicts and generates immunogenic peptides to assess T cell antigen recognition and support cancer immunotherapy and vaccine design.

Key Features:

• Predictive Modeling: Employs a convolutional neural network (CNN) to predict peptide immunogenicity from sequence data and outperforms ElasticNet, KNN, SVM, Random Forest, AdaBoost, ResNet, and GNN in reported benchmarks.
• Benchmarking: Validated on three independent immunogenic peptide collections derived from dengue virus, cancer neoantigens, and SARS-CoV-2 and compared against DeepHLApan and IEDB.
• Epitope Importance Prediction: Identifies critical peptide residue positions associated with T cell antigen recognition.
• Generative Capabilities: Implements DeepImmuno-GAN, a generative adversarial network that simulates immunogenic peptides with physiochemical properties and predicted immunogenicity similar to real antigens.
• Immunogenicity Scoring: Uses a beta-binomial distribution approach to derive epitope immunogenic potential from peptide sequences.

Scientific Applications:

• Cancer Immunotherapy: Prediction of non-native and neoepitopes to inform targeted T cell–based cancer therapies.
• Vaccine Development: Identification and simulation of candidate immunogenic peptides for vaccine antigen selection against pathogens.
• HLA Allele Simulation: Generation and prediction of peptides tailored to specific human leukocyte antigen (HLA) alleles to support personalized immunotherapy and vaccination strategies.
• Synthetic Peptide Design and Dataset Augmentation: Use of GAN-simulated peptides to support synthetic biology applications and augment training datasets.

Methodology:

Uses a convolutional neural network (CNN) for immunogenicity prediction, a generative adversarial network (DeepImmuno-GAN) for peptide simulation, and a beta-binomial distribution to derive epitope immunogenic potential; benchmarking employed peptide collections from dengue virus, cancer neoantigens, and SARS-CoV-2 and comparisons to ElasticNet, KNN, SVM, Random Forest, AdaBoost, ResNet, GNN, DeepHLApan, and IEDB.