Sajib Acharjee Dip, Adrika Zafor, Bikash Kumar Paul, Uddip Acharjee Shuvo, Muhit Islam Emon, Xuan Wang, Liqing Zhang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Antibiotic resistance is a silent pandemic, causing 700 thousand human deaths across the world every year. Antibiotic resistance genes (ARG) are genes conferring resistance for the bacteria carrying them. Predicting ARGs is an important computational task. Traditionally ARGs are predicted using alignment based methods. However, the false negative rate for most of the alignment-based tools is very high. The protein language models (LM) trained on the huge corpus protein sequences capture distant relations among protein sequences. These features can be utilized for the identification and classification of ARGs. We have presented a self-supervised model on the largest available ARG database with the help of a pre-trained language model ProtAlbert. We used the raw protein LM-embeddings from unlabeled data on our ARG classification task and saw it outperform state-of-the-art prediction algorithms. The extracted features from the pretrained language model boosted the supervised model accuracy to a great margin.