Trustworthy AI for DNA Sequencing

We live in an era of data abundance with ever growing storage requirements. DNA has become a desirable storage medium due to its durability, density and high storage capacity. DNA sequencing is at the heart of DNA storage. It has been successfully applied to domains outside data storage, including precision and personalised medicine. DNA sequencing is a process whereby the order of the DNA nucleotides is determined. Artificial Intelligence (AI) techniques, deep learning in particular, have largely benefited DNA sequencing, making the signal mapping from the sequencer to nucleotides or to the whole sequence cheaper and faster. Despite this advancement, DNA sequencing is still hampered by a large error rate and the lack of output explainability, i.e. how a series of signals is mapped to a sequence (Data Storage in DNA). For medical applications, this is a dangerous setback as it might lead to adverse effects on patients and increased medical costs. The limitations associated with the sequencing process also affect DNA storage. Various studies have attempted to address high error rates by improving on sequencing technologies or error correction algorithms. However, error rate reduction is still unsatisfactory. In addition, the question of how a DNA sequencing result was obtained remains unanswered. This study takes at its premise the view that these limitations must be tackled from a different angle. Namely, applying rule-based logic to the sequencing process.

OBJECTIVES

This study builds on the Oxford Nanopore Technologies (ONT) software to explore how Probabilistic Logic Programming (PLP) can be used to create logic-based rules for improving the high error rate and traceability problems associated with DNA sequencing. A macro-level objective is to enable trust in DNA sequencing. On a micro level, the aim is to explore how PLP can be utilised for DNA storage. This raises the following questions:

- Can a single PLP model replace all current ONT ML models? How...