COVID‑19 Modelling & Epidemiological Analysis

During the COVID‑19 pandemic, I helped lead of a Royal Society initiative

to find and deploy computer-literate volunteers to assist modelling the epidemic. Science under time-pressure is very different from normal research: a good model now is better than a perfect model too late.

Benchmarking and redteaming

The most influential model was Covidsim from Imperial College. We (EPCC) parallelised and ported this model to Archer, the national supercomputer and demonstrated its reproducibility

demonstrated its reproducibility.

It was a bizarre time: an irrelevant bug in initialising the random number seed was raised by David Davis MP in the House. Because we corresponded with ICL, our work was deemed "not independent". Although the code was correct, the "Report 9" which was the prime model used to justify the lockdown suggested that school closures would result in more deaths (as well as educational disruption). It also forecast that more people would die if there was a lockdown, although there were so many unknowns here we did not promote this point. All this was in the public domain in "Report 9". Nobody reads the facts. I had my 15 minutes of fame of the Today programme, was praised in the anti-lockdown Telegraph, and lambasted in the pro-lockdown Guardian.

Weight, Scale and Shift model

Within RAMP, I was leading the "New Modelling" initiative. Scientists love making models and hate data-wrangling. This was meant to assess the flood of new pandemic models. There was a lot of nonsense, several groups who had credible but biassed models focussed on publicity rather than verification. Two initiatives really stood out as superior software: PyRoss (Cambridge) and JUNE (Durham). Then came WSS.

Weight, Scale, Shift

EPCC founder David Wallace emailed to say he had a model for predicting the R-number. It was in excel. It got the same predictions as the official SAGE data published weekly. And it got them two weeks earlier.

WSS Early evidence: Alpha (B.1.1.7) was more deadly

The Alpha variant emerged from Kent in late 2020. Very early data from NHS England suggested it was no more deadly. Our analysis showed otherwise. We published our data on medarXiV on Monday. On Thursday the Prime Minister announced that B1.1.7 was indeed more deadly.

Dynamics and oscillatory solutions

Subsequent theoretical work in Journal of Theoretical Biology explored stability and oscillations in epidemic systems, informing how interventions may create non‑intuitive dynamics.

Understanding model reliability

A review of forecasting approaches showed that many publicised pandemic models under‑performed baselines and understated uncertainty, motivating transparent, testable frameworks.