PhD studentship in Explainable Network and AI-based methods for personalised multi-omic medical data analysis

Overview

Qualification type: PhD

Location: Hatfield, United Kingdom

Expected start: October 2021

Project outline

Recent technological developments allow us to collect a tremendous amount of digital personal multi-omic medical data that include clinical parameters, serum and urine biomarkers, or genomic data. Some of these analytes are collected over time and organise serial or longitudinal data. Resulting datasets are frequently characterised by small sample sizes and large numbers of features. This results in that fact that existing machine learning approaches often fail to analyse such data and are prone to overfitting.

One of the ways to analyse multiple features from patients in such setting is to use a network approach such as parenclitic networks [1,2]. Although it was shown to be promising in analysing low sample size multidimensional datasets, there are still unsolved methodological issues such as: i) which topological indices are most relevant for predictive modelling purposes; ii) what are the different network architectures; iii) how to incorporate longitudinal characteristic into the network setting; and, finally, iv) how to make these models transparent and explainable. This latter is crucial in order to be able to interpret such black box approach and identify which particular patterns in the data have led to a specific classification result. In the present project we plan to develop a network and AI methodological approaches to challenge these questions.

[1]. H.J. Whitwell, O. Blyuss, J.F. Timms, and A. Zaikin, “Parenclitic networks for predicting ovarian cancer”, Oncotarget  9:32, 22717-22726 (2018).

[2]. A. Karsakov, T. Bartlett, I. Meyerov, M. Ivanchenko, and A. Zaikin,  “Parenclitic network analysis of methylation data for cancer identification”, PLOS ONE 12(1), e0169661  (2017).

[3]. V. Demichev, P. Tober-Lau, T. Nazarenko, C. Thibeault, H. Whitwell, et al. “A time-resolved proteomic and diagnostic map characterizes COVID-19 disease progression and predicts outcome”, medRxiv, doi 10.1101/2020.11.09.20228015 (2020).

Supervisors

Dr Oleg Blyuss and Prof Jim Geach. Email enquiries to Dr Oleg Blyuss o.blyuss@herts.ac.uk.

Essential criteria

Students are normally required to have at least a 2.1 undergraduate degree in a relevant subject (or an international equivalent), and we also expect applicants to have an IELTS (English proficiency) score of normally 6.5 or above.

Application process

To apply, a completed application form, along with copies of higher education certificates, transcripts, a copy of your passport and two letters of reference, should be returned (by post or email) to Lynette Spelman at the address below, by the closing date. You should arrange for your referees to write separately using the form provided or send their reference via email to Lynette Spelman.

Download the application form (PDF - 0.25 Mb).