Postdoc in Synthetic Data Simulation for Malaria Genomic Epidemiology

Institut Pasteur

Paris, France

Using simulation-based inference (SBI) with deep learning, UniGEM aims to build a neural network to estimate epidemiological parameters of P. falciparum, the deadliest malaria parasite species. Within UniGEM, the postdoc will be responsible for the simulation of synthetic data used to train, develop, and test the neural network.

Scientific context: Malaria parasites are transmitted between humans by mosquitoes. In the mosquito host, they undergo an obligate stage of sexual recombination. Transmission-dependent rates of out-crossing, inbreeding, and self-fertilization between parasites are governed by superinfection and co-transmission on the mosquito-to-human level. These overlapping, non-linear processes obfuscate estimation of malaria epidemiological parameters from parasite genetic data; cf., viral genomic epidemiology, where bifurcating trees capture the ancestry of DNA sequences, and human population genetics, where the ancestral recombination graph (ARG) needs no dynamic rate of self-fertilisation or hierarchical structure over processes on a host-to-host level. To estimate malaria epidemiological parameters using simulation-based inference, UniGEM requires a simulator that captures malaria-specific recombination and transmission processes, generating realistic synthetic datasets while retaining sufficient computational eXiciency for large-scale simulation.

Postdoc work programme:

The postdoc will review the literature on epidemiological parameters and existing simulators, then benchmark a subset of simulators against key criteria. They will likely iterate twice over data simulation: once towards the end of their first year, using the most appropriate simulator available at that time. A second year provides time to develop an enhanced simulator and to simulate a refined batch of synthetic data. For each batch, the postdoc will estimate Bayes optimal error, an important guide for realistic goals for deep learning. The first batch of synthetic data will be used by a PhD student, likely starting October 2027, to initiate development of a deep-learning architecture. The refined batch of synthetic data will be used by the PhD student to finalise the deep-learning model. Expected outputs include parameter priors, benchmarking results, a simulation pipeline, curated synthetic datasets, and Bayes optimal errors.

Publication/conference:

Depending on progress, there is potential for one or two first-author publications: a benchmarking study comparing existing simulators and a methodological study detailing the development of an enhanced simulator. In addition, the postdoc will be invited to co-author all publications on UniGEM activities that make use of the postdoc results and encouraged to disseminate findings at the annual American Society of Tropical Medicine and Hygiene conference.

Candidate profile:

An ideal candidate will likely have a PhD in statistical population genetics (e.g., prior experience working with ARGs/coalescent-with-recombination), computational biology, bioinformatics, infectious-disease epidemiology, evolutionary or ecological biology, or a closely related field.

• Strong programming skills (R, Python, and/or other)
• Familiarity with version control (Git/GitHub)
• Proficient written and spoken English (B2 level or higher)
• Ability to communicate methodological assumptions and limitations clearly and succinctly
• Experience with reproducible computational workflows and high-performance computing
• Experience handling large population-genetic simulations (e.g., SLiM, msprime, tskit)