Researchers Xinru Wen, Weizhong Lin, Zi Liu, and Xuan Xiao from the School of Information Science and Engineering at Yanshan University have developed a novel approach to identify antiviral peptides (AVPs), which could significantly impact drug development in the energy industry’s fight against viral infections that can disrupt operations and endanger workers.
The team’s work, titled “AVP-Pro: An Adaptive Multi-Modal Fusion and Contrastive Learning Approach for Comprehensive Two-Stage Antiviral Peptide Identification,” addresses the challenges of capturing complex sequence dependencies and distinguishing between similar samples. Their solution, AVP-Pro, is a two-stage predictive framework that integrates adaptive feature fusion and contrastive learning. This approach aims to improve the accuracy of identifying AVPs, which are crucial for developing novel antiviral drugs.
In the first stage of AVP-Pro, the researchers constructed a panoramic feature space encompassing 10 distinct descriptors to capture the physicochemical properties and deep-seated patterns of peptide sequences. They designed a hierarchical fusion architecture that integrates self-attention and adaptive gating mechanisms. This architecture dynamically modulates the weights of local motifs extracted by convolutional neural networks (CNNs) and global dependencies captured by bidirectional long short-term memory networks (BiLSTMs) based on sequence context.
To tackle the issue of blurred decision boundaries caused by the high similarity between positive and negative sample sequences, the researchers adopted an Online Hard Example Mining (OHEM)-driven contrastive learning strategy enhanced by BLOSUM62. This approach significantly sharpened the model’s discriminative power. In the first stage of general AVP identification, the model achieved an accuracy of 0.9531 and a Matthews correlation coefficient (MCC) of 0.9064, outperforming existing state-of-the-art methods.
In the second stage of functional subtype prediction, the researchers combined a transfer learning strategy with AVP-Pro to accurately classify 6 viral families and 8 specific viruses under small-sample conditions. This capability is particularly valuable for the energy industry, as it enables high-throughput screening of antiviral drugs tailored to specific viral threats.
The researchers have developed a user-friendly web interface for AVP-Pro, making it accessible for users in the energy sector and beyond. This tool provides a powerful and interpretable means of identifying AVPs, which can be crucial for developing antiviral drugs to protect workers and maintain operations in the energy industry.
The research was published in the journal Briefings in Bioinformatics, a reputable source for bioinformatics research. The findings highlight the potential of advanced machine learning techniques in improving the accuracy and efficiency of antiviral drug development, offering practical applications for the energy industry’s health and safety efforts.
This article is based on research available at arXiv.