Sam Abbott

grant proposal actual research

Supporting exciting biological sciences ideas and impacts on AI and biotechnology

If anyone is interested in writing an NSF PRFB applications on the intersection of AI, disease forecasting, and understanding drivers of transmission (or related topics), please reach out! I think it would be a great fit with the work happening in our lab: www.nsf.gov/bio/updates/...

Thanks @sbfnk.bsky.social, yes there does seem to be some traction progress in a few places which is great news.

Great talk on composable infectious disease modelling. Making real-time outbreak modelling more principled and flexible has been a long-standing challenge, and Sam has been leading the charge on solving it. With a growing group of people keen on this it feels like we're finally getting somewhere.

Design and analysis to estimate diagnostic test accuracy without universal verification at University of Bristol on FindAPhD.com PhD Project - Design and analysis to estimate diagnostic test accuracy without universal verification at University of Bristol, listed on FindAPhD.com

📣 PhD studentship - Biostatistics 💻
University of Bristol, UK
Help improve methods for something that affects us all - how accurate are tests for disease?
Learn, apply & develop Bayesian models to estimate accuracy from varying study designs, synthesise evidence & optimise design of future studies

Thanks! Totally agree.

I think composable models are the future, and it would be great to have more tools for building composable models. This is a nice talk about the benefits

hmmTMB: Hidden Markov Models with Flexible Covariate Effects in R by Théo Michelot <p>Hidden Markov models (HMMs) are widely applied in studies where a discrete-valued process of interest is observed indirectly. They have for example been used to model behavior from human and animal...

The hmmTMB paper is finally out in the Journal of Statistical Software!

An R package for hidden Markov models with random effects, flexible spline-based covariate effects, and fast inference using TMB or Stan.

Check out the GIthub repository for more examples.

doi.org/10.18637/jss...

The Tyranny of the Marginal User why consumer software gets worse, not better, over time

nothinghuman.substack.com/p/the-tyrann...

Being popular causes all kinds of bad context collapse phenomena

Really cool that @kaitejohnson9.bsky.social is giving a keynote at the next Stan conference!

How the processes of recuperation vitiate apparent methodological advances in medical research, from meta-analysis to Mendelian randomization

As ever I was actually thinking of some specific examples here but hard to nicely discuss or study at the individual level

You would have so major censoring issues (as the real extreme casss won’t show up) but presumably you could do a study on this using lab GitHub orgs and some measure of output quality (always very tricky)

On the other hand maybe the vibing off of graphs skills are what we need in the age or perhaps llms improve code quality in these labs as those doing the code are usually pretty junior/untrained with little to no positive reinforcemen.

I have always thought that is a serious issue as it leaves so much room for errors and ambiguity. This would mean with llms we might expect reduced study quality from these folks.

The example I am thinking about here is the stereotype mid to senior academic who hasn’t looked at code in a decade and judges studies by vibing off of the graphs and the high level summaries.

My dog walk thought for the day: LLM coding tools must hit very differently for different levels of seniority(kind of obvious)

🤩 unfortunately it doesn’t include the discussion which I was pretty good but I think there’ is going to be a follow up.

Posterior-SBC now also with peer-review stamp in Statistics and Computing doi.org/10.1007/s112... (update your bib files)

Peer review gets a lot of kudos, but often it's the feedback we get direct from people we know and trust before a paper is even submitted that feels like it has the most significant impact on our work. These people usually show up in the acknowledgements, where they get almost no recognition at all.

If you want to read more about the work I presented on see: epiaware.org/ComposablePr...

The Juniper seminars are we well worth attending. See juniper.maths.org/seminars

YouTube video by Juniper Consortium Seminars Sam Abbott (LSHTM): Composable probabilistic models can lower barriers to rigorous ID modelling

I had a great time talking at the Juniper seminar series last week about composable infectious disease models. Some very good discussion after the talk. The recording is now up!

youtu.be/FQYOqGnbJWA?...

🦹‍♂️ success

Slides and context for this here.

community.epinowcast.org/t/new-pre-pr...

Please be nicer about my figures than @josephlemaitre.bsky.social was as yes they don't look great but I tried hard 😢😉🥳

Haha good to hear it!

I like modernjuliaworkflows.org a lot.

I also have a little list of stuff here: github.com/EpiAware/Cen... (github pages seems to be having problems!

There is some #julia stuff on here but I would also check out the julia discourse and slack both nice places

Claude and I are going to have lots of fun with this!

Thanks yes makes sense. In r land the standard is quarto which gives you both a notebook and running which I like but I do like the idea of literate being very light weight.

yes fair enough that makes sense.

*figure code

A conceptual figure outlining a protocol for using species distribution models to prioritize sampling for pathogens in wildlife

A scientific figure showing a case study of the protocol applied to Arenaviridae in rodents in India

🚨🐭🦠 New preprint!

Monitoring wildlife disease is expensive, so we need to be smart about where we sample to get as much useful info as possible.

We show how SDMs can guide adaptive sampling for zoonotic disease surveillance, using case studies in rodents.