AI3SD Project: Application of Capsule Net for automated DNA sequencing using tunnelling spectroscopy

Abstract

In this project, we take the next step for quantum tunnelling-based biosensing and sequencing to become promising contender for label-free ‘next-next’ generation sequencing of biopolymers.This could be achieved by combining state-of-the-art surface chemistry, nanoscience and chemical sensing with Capsule Nets (CN) as a novel Deep Learning methodology, to maximise the extraction of information from the tunnelling data. Our results to date show a moderate improvement in detection accuracy when using Capsule Nets, compared to Convolutional Neural Networks. CapsNets were however superior when dealing with incomplete data, for example
when recognising events at the edges of the recording window, which could partly offset the increased computational cost. Ongoing work beyond this project now aims to further explore this aspect and support our findings with increasing amounts of experimental data. In addition, the project has allowed us to explore a new approach to data classification of single-molecule charge transport data using Transfer Learning and Image Recognition Networks (IRN). In this context, the feature extractor of IRNs trained on millions of (unrelated) image data is used to recognise characteristics in conductance-distance or current-time data, which are then clustered and interpreted. This means that no charge transport data are required for training the feature extractor, removing the need for large, application-specific training data.

More information

Identifiers

ORCID for Samantha Kanza: orcid.org/0000-0002-4831-9489

ORCID for Jeremy G. Frey: orcid.org/0000-0003-0842-4302

Catalogue record

Export record

Altmetrics