Feature engineering (FE) is highly crucial in machine learning (ML), which directly governs a predictive model's performance and thereby serves as a foundational step for accurate and reliable biomolecular recognition. However, most existing ML-based sequencing approaches rely predominantly on conductance and current-based descriptors, and the absence of a well-defined molecular fingerprint highlights a pressing need for more informative FE. So, intending to expand the feature landscape, herein, we propose an ML-driven tool to address the challenges associated with the current ML-aided sequencing methods. Our approach enables the construction of a more robust feature space by integrating both conventional quantum transport descriptors and newly incorporated electronic density of states (DOS)-based descriptors. While the transmission signals encode the intrinsic transport properties of target molecules, the DOS provides key insights into the local electronic structure and the extent of electrode-molecule coupling, serving as a rich source of highly informative features. On implementing our idea, the proposed ML classification approach successfully distinguishes all the biomolecules (four DNA nucleotides as well as six structurally diverse blood antigens with regioisomeric glycosidic linkages), achieving an accuracy of up to 100%. This study, therefore, provides proof of concept that DOS-based descriptors are as reliable as transmission-based features for biomolecular recognition while offering a simpler experimental alternative.
Chatterjee et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: