Rachel's Interview & Publication

SYNLAB UK & Ireland

Dr Rachel S. Carling

Director of Newborn Screening and Clinical Lead, Biochemical SciencesScientific Director, Viapath (A partnership between the NHS and SYNLAB UK & Ireland)
Guys & St Thomas’ NHS Foundation Trust, London

A Machine Learning Approach for the Automated Interpretation of Plasma Amino Acid Profiles

Clinical Chemistry – https://doi.org/10.1093/clinchem/hvaa134

What inspired your research and what did it cover?

An initial project demonstrated proof of concept, which is fantastic, but for this to improve the quality and efficiency of the laboratory service we provide, it needs to be translated into routine use. My focus has been on helping to ‘operationalise’ the algorithm. The goal is to create a simple PoC web interface through which we would upload our amino acid results from the mass spec, marry them up with key demographic information from the LIMS [Laboratory Information Management Systems], and have the ML [Machine Learning] model’s inference output delivered back to the web interface in real time.

Which aspect(s) of your research work are you particularly excited about?

The outlook of applying machine learning to other diagnostic tests, both within my area (e.g., acylcarnitine and organic acids) and more widely across medical diagnostics.

Looking at the potential of your findings, what difference can they make?

There is great potential in their wider application for personalized medicine in diagnostics, and specifically in the ways they can benefit

  • Clinical practitioners: Multi-centre international collaboration is planned with other laboratories. This will allow us to really test the algorithm by exposing it to a much larger number of results and diagnostic cases, and to determine if a single algorithm can be used by different methodologies and laboratories.
  • Patients: Faster turnaround time and reduced potential for a missed diagnosis. Many of these disorders are very rare (1 in 100,000) and scientists may never see a case in their lifetime. An algorithm that has been trained and continues to learn would be very powerful.
  • Customers: If successfully operationalised, ML can increase operational efficiency and process quality to improve throughput. It further addresses the acknowledged shortage of scientific expertise in this area.