A Computational Model for Identifying Antigenic Proteins using Deep Learning

Abstract

Accurately identifying antigenic proteins is essential to immunology and vaccine development. Experimental laboratory methods are accurate but costly, time-consuming, and resource-intensive. Computational methods, particularly those that use Deep Learning, offer a quick and inexpensive substitute for antigenicity prediction. In this work, we developed a deep learning-based compu tational model that classifies protein sequences as either antigenic or non-antigenic. We trained different deep learning frameworks including Gated Recurrent Network (GRU), Bidirectional Long Short-Term Memory (BiLSTM), and Multiheaded Convolutional Neural Network (MH-CNN). The features are explored by protein feature extraction methods such as UniRep embeddings, ProtBert, and Evolutionary Scale Modeling (ESM). Our comparative evaluation revealed that the MH-CNN with UniRep embeddings achieved the highest performance. Users can input protein sequences and receive immediate predictions about their antigenicity through a web application built with HTML, CSS, and Flask. This method significantly reduces the time and expense of antigen discovery, making it a valuable tool for researchers.