DeepSME: De Novo Nanopore Basecalling of Motif-insensitive DNA Methylation and Alignment-free Digital Information Decryptions at Single-Molecule Level

ABSTRACT

Biomolecular encryption employing chemical modifications enables secure approaches for information storage and communications. However, constructing high information density pathways for rapid synthesis and readout remains a challenge to guarantee confidentiality, integrity, and availability (CIA). Here we develop a nanopore sequencing based protocol, demonstrated by complete substitution using 5-hydroxymethylcytosine (5hmC) for individual nucleotide recognition rather than sequential interactions. Such motif-insensitive methylation at the single-molecule level does not naturally exist and results in severe ion current disruption and a 67.2%-100% readout failure, which ensure its ability on the encryption of the data encoded inside the DNA. We further propose and establish an alignment-free DeepSME basecaller, which is a deep learning-based platform independent on prior models and knowledges. DeepSME utilizes a three-stage training pipeline that initiates tolerable for 11.55% errors, expands its neighboring k-mer dictionary model size from 4^6 to 4^9, and mitigates the errors by only three microbial genomes, giving rise to 92% precision with 92% recall. Fully 5hmC encrypted digital information were deciphered by DeepSME within 16× coverage depth. The versatile and transparent DeepSME pipeline and its F1-score performance of 86.4% surpassing all the state-of-the-art basecallers, support its great potential for meeting the rapidly increasing CIA demands of DNA-based secure communications.