Machine learning prediction and classification of transmission functions for rapid DNA sequencing in hybrid nanopore

Abstract

Electrical DNA sequencing using solid-state nanopores has emerged as a promising technology due to its potential to achieve high-precision single-base resolution. However, uncontrollable nucleotide translocation, low signal-to-noise ratios, and electrical signal overlapping from nucleotide stochastic motion have been major limitations. Recent fabrication of in-plane hybrid heterostructures of 2D materials has triggered active research in sequencing applications due to their interesting electrical properties. Herein, our study explores both machine learning (ML) regression and classification framework for single DNA nucleotide identification with hybrid graphene/hexagonal boron nitride (G/h-BN) nanopore using a quantum transport approach. The optimized ML model predicted each nucleotide at their most stable configurations with the lowest root-mean-squared error of 0.07. We have also examined the impact of three locally polarised hybrid nanopore environments (C𝛿−−H𝛿+, N𝛿−−H𝛿+, and B𝛿+−H𝛿−) on ML prediction of transmission functions utilizing structural, chemical, and electrical environmental descriptors. The random forest algorithm demonstrates notable classification accuracy across quaternary (~86%), ternary (~95%), and binary (~98%) combinations of four nucleotides. Further, we checked the applicability of the hybrid nanopore device with conductance sensitivity and frontier molecular orbitals analysis. Our study showcases the potential of a hybrid nanopore with ML combined quantum transport method as a promising sequencing platform that paves the way for advancements in solid-state nanopore sequencing technologies.