Bayesian inference and AI approaches to infer patterns of disease spread from the genomes of the malarial parasite

Project description

Understanding spatial connectivity is critical to malaria control: otherwise human communities can be continually seeded with new infections. Traditional methods for identifying infection routes rely on patchy information about human travel. Genomic data, from the malarial parasite itself, offer a far more direct and powerful information, but the analysis must account for background levels of relatedness within parasite populations.

A fundamental problem is that allowing for relatedness involves partitioning the very large number of parasite genomes into different categories, depending on a model’s assumptions about the network of connections across the landscape. The multiplicity of models makes a likelihood-based analysis prone to misinterpreting the data, by failing to find the well supported models among the profusion of less effective candidates.  To overcome this challenge, the student will implement a solution borrowed from physics – the use of parallel tempering, which uses linked MCMC chains running at a range of temperatures, whereby high temperature chains scrutinise the prior range of models, whilst the lower temperature chains explore the better candidates. Efficient implementation will be achieved using an engine for Bayesian inference by parallel tempering, maintained by a previous PhD student of RN (https://mrc-ide.github.io/drjacoby/). Synthetic data, and likelihood functions will be developed in collaboration with Alexander Gnedin (SMS), an expert on the mathematics of the required lambda coalescent models. This broad strategy will be compared with the use of Generative Adversarial Networks – supervised AI learning methods that can to both generate both artificial genomes and estimate cryptic population genetic parameters, which will be implemented with the support of Matteo Fumagalli (SBBS).

The overall aim is to analyse real Plasmodium falciparum datasets, to visualize and interpret their patterns for malarial control. The student will obtain expertise in two fundamentally distinct (Bayesian & GAN) cutting-edge approaches to solving the big data challenges of genomic data.

Funding

This studentship is funded by QMUL and open to UK students. It will cover tuition fees, and provide an annual tax-free maintenance allowance for 3 years at the Research Council rate (£17,609 in 2021/22).

Eligibility and applying

Applications are invited from outstanding candidates with or expecting to receive a first or upper-second class honours degree and a masters degree in a biological discipline and / or area related to mathematics and / or statistics and / or data science. This is to be understood in a broad sense, please get in touch for an informal discussion by writing an e-mail to [Email Address Removed]

Additional skills required:

• An appetite for coding and mathematical modelling to solve problems in evolutionary genetics is more important than extensive experience.

Applicants from outside of the UK are required to provide evidence of their English language ability. Please see our English language requirements page for details.

Formal applications must be submitted through our online form by the stated deadline including a CV, personal statement and qualifications.

The School of Biological and Behavioural Sciences is committed to promoting diversity in science; we have been awarded an Athena Swan Silver Award. We positively welcome applications from underrepresented groups.

http://hr.qmul.ac.uk/equality/ 

https://www.qmul.ac.uk/sbbs/about-us/athenaswan/

Apply Online