Publication

Abstract : Low Complexity Regions (LCRs) plays a significant role in the cellular mechanisms of proteins, such as evolutionary processes and the regulation of protein activity. These regions are characterized by repetitive amino acid sequences and varied subtle properties, making LCR identification a computational challenge. Traditional methods for LCR detection, while informative, are labor-intensive and lack the scalability and specificity required for rapid analysis across exponentially growing protein sequence databases. Addressing these limitations, the paper proposes a deep learning framework that utilizes a pre-trained BERT model and K-mer analysis for feature extraction, combined with a BiLSTM network, to precisely predict LCRs within protein sequences. By adapting advanced natural language processing techniques to the unique challenges of protein sequence analysis, our approach significantly outperforms other machine-learning models, as evidenced by an MCC of 0.2454 and an AUC of 0.8627. This advancement opens new avenues for understanding protein function and pathology, offering a robust tool for the accelerated discovery of protein dynamics elucidation.