Amino acid sequence-based IDR classification using ensemble machine learning and quantum neural networks

Biologically traditional methods, such as the Uversky plot, which rely on hydrophobicity and net charge, have inherent limitations in accurately distinguishing intrinsically disordered regions (IDRs) from ordered protein regions. To overcome these constraints, we propose a novel ensemble framework integrating Machine Learning (ML), Deep Neural Networks (DNN), and Quantum Neural Networks (QNN) to enhance IDR classification accuracy. Notably, this study is the first to employ QNNs for IDR classification, leveraging quantum entanglement to model intricate feature interactions. Amino acid sequences were analyzed to extract biophysical features, including charge distribution, hydrophobicity, and structural properties, which served as inputs for the predictive models. ML was utilized for independent feature learning, DNN for hierarchical interaction modeling, and QNN for capturing high-order dependencies. Our meta-model demonstrated an accuracy of 0.85, surpassing individual classifiers and highlighting the importance of buried amino acids and feature interactions between scaled hydrophobicity and large, buried, and charged residues. This study advances computational protein science by demonstrating the applicability of QNNs in bioinformatics and establishing a robust framework for IDR classification.