New pre-pre-print: Composable probabilistic models can lower barriers to rigorous infectious disease modelling

Publicity
1

Visual Abstract
Visual Abstract1920×1059 438 KB

Paper site: https://epiaware.org/ComposableProbabilisticIDModels/
Paper: https://epiaware.org/ComposableProbabilisticIDModels/paper.html
Case studies: https://epiaware.org/ComposableProbabilisticIDModels/case-studies.html

The visual abstract is slightly mad at points but I haven’t had the courage to return Nano Banana pro to do battle yet.

Very, very keen to hear feedback and what people think about some of the ideas in here.

TLDR: Disease outbreak models need to be fast, accurate, and built collaboratively, but current approaches force a trade-off. You can either chain simple models together (flexible but loses information) or build one big custom model (rigorous but not reusable). This paper proposes a middle ground: modular building blocks that snap together and can be reused across projects. We built a proof of concept in Julia/R and showed it can replicate published COVID-19 and influenza analyses by mixing and matching components. The approach makes it easier for different experts to contribute pieces and could work well with AI coding assistants.

1 Like
2

This is neat, love the framing. I’ve shared that with some of our team on the model building side. Are the little flow graphs also nano-banana ? These are really not to my taste :smiley: (maybe better to ask it to generate svg ?)

1 Like
3

The ones in the paper? They are mermaid ( ComposableProbabilisticIDModels/figures/fig-case-studies.qmd at d9d7f7737335dc048ef611bba3b57d7a6d296a6f · EpiAware/ComposableProbabilisticIDModels · GitHub ) and it would be nice to get svg out of possible but I haven’t worked out how to.

Do you have a preferred tool for making these sorts of diagrams? I need a good one but ideally one that is programmable.

4

mmh not I would like to know a programmable one I always oscillate between draw.io (it’s xml so should be editable) and others.

yeah meant figure 1 here: Composable probabilistic models can lower barriers to rigorous infectious disease modelling

1 Like
5

I gave a talk on this to Juniper for which I think there will be a recording shortly. In the meantime some slides are here:

6

Recording of my recent Juniper talk on this work: https://youtu.be/FQYOqGnbJWA?si=VHbZAvv3RiEFdkb2