A Weighted Deep Learning Approach to Identify Nucleic Acid-Binding Proteins
Authors :- Mahala, A., Ranjan, A., Priyadarshini, R., Dansena, P., Tripathi, P.C.
Publication :- Proceedings of International Conference on Computational Intelligence: ICCI 2024, Springer, 2025
Nucleic-acid binding proteins (NABPs) are proteins that interact with nucleic acids such as DNA and RNA to conduct various cellular-level processes. Therefore, to strengthen and fasten the knowledge about NABPs, simple and efficient computational methods are much needed. The work introduces a sub-sequence-level architecture, i.e., weighted deep GRU + multi-attention model, to solve the NABPs identification task. GRU + multi-attention can capture temporal relationships between amino acids while emphasizing the most informative features for classification. Further, the weighing scheme helps address the issue of data imbalance. The experimental results by employing three independent test datasets suggest a great potential with the proposed architecture. The weighting approach produced improved results across both classes, including the minor class.