Vaxi-DL: A web-based deep learning server to identify potential vaccine candidates

Area of Science:

• Computational biology
• Vaccine development
• Bioinformatics

Background:

• Vaccine development is a lengthy process involving multiple stages.
• In silico screening aids antigen identification to accelerate vaccine design.
• Current prediction tools require improvement for accuracy and efficiency.

Purpose of the Study:

• To introduce Vaxi-DL, a web-based deep learning (DL) software for evaluating protein sequences as vaccine target antigens.
• To develop and validate four DL models for predicting target antigens across bacteria, protozoa, fungi, and viruses.
• To benchmark Vaxi-DL against existing prediction tools and assess its performance on known vaccine candidates.

Main Methods:

• Trained four DL models using datasets of antigenic and non-antigenic sequences from vaccine candidates and the Protegen database.
• Computed biological and physicochemical properties, then divided datasets into training, validation, and testing subsets.
• Constructed models using Fully Connected Layers (FCLs), hyper-tuned, and assessed performance using metrics like accuracy, sensitivity, specificity, and AUC.

Main Results:

• Vaxi-DL demonstrated high performance across four pathogen models, with an average sensitivity of 93%.
• The tool correctly predicted 175 out of 219 known potential vaccine candidates (PVCs) from 37 pathogens.
• Benchmarking showed Vaxi-DL comparable or superior performance to existing tools like VaxiJen and Vaxign-ML.

Conclusions:

• Vaxi-DL is a valuable deep learning tool for prioritizing potential vaccine candidates for preclinical studies.
• The software accelerates antigen identification, contributing to more efficient vaccine development pipelines.
• Vaxi-DL's high accuracy and broad applicability make it a significant advancement in vaccinology.