iiinova — where your digital self keeps evolving

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

I should have said at the top: this paper has now been accepted at Philosophical Transactions of the Royal Society B, as part of a forthcoming special issue on brain-body interactions in health and disease! In the meantime, the author accepted manuscript is available on PsyArXiv (see link at top)

010

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

And finally of course my postdoc supervisor @camillanord.bsky.social (who actually originated the idea and had already been testing with an early version of the study when I arrived in her lab) who has been so supportive over the years!

110

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

@gabmackie.bsky.social @isabellau.bsky.social and Rukmini Chariar. Also grateful to collaborators who've lent resources and given other support to the project – Tim Dalgleish, Anna Bevan and @alpowers7.bsky.social.

120

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

This study has been a massive effort over a number of years (it was the first thing I worked on in my postdoc!) and has involved many people in that time. Massive thanks to the RAs and students who helped not just with data collection but also just getting the paradigm to work in the first place –

110

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

In summary: the new task provides a valuable method for looking at how somatic symptoms may be generated, and we uncovered some initial evidence that prior beliefs (and possibly how they are updated over time) do indeed differ in anxiety.

110

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

Instead, it seems that participants in the high anxiety group held stronger beliefs (μ2 in the model), and maintained these for longer despite the changing evidence (though the latter result only emerged in sensitivity analyses - so take with a pinch of salt!)

Plots showing parameters from the HGF (mu_2 - the strength of the prior belief, and sigma_2 the uncertainty of this belief)

110

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

What about the computational mechanism? For this, we fitted an HGF and, contrary to our initial hypothesis, there was no difference in ν, a parameter describing the weight given to the prior vs sensory evidence in the decision-making phase (and which had shown up in Powers et al.'s earlier paper)...

110

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

Why? It wasn't a sensory difference: perceptual sensitivity (d') declined over time by the same amount for both groups. Instead, it was their response criterion that diverged – the high anxiety group maintained a comparatively more liberal threshold for longer.

Graph showing response criterion increasing over time by more for the low anxiety group (blue) than high anxiety group (red)

120

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

YES – all participants reported a significant number of 'phantom percepts' (detection when the stimulus was absent). In addition, the high anxiety group maintained a higher overall detection rate as the stimulus contingency fell over time, whereas the low anxiety group adapted more quickly.

Graph showing the changing detection rates of the high (red) and low blue) anxiety groups. Over the course of the task, the tactile stimulus was less and less likely to actually be delivered; low anxiety patients were increasingly less likely to report feeling the stimulus, while those in the high anxiety group maintained higher detection rates for longer.

110

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

Adapting Powers et al.'s auditory paradigm to somatosensation, a visual cue was repeatedly paired with a mild electrical stimulus at the ankle. Over time, the stimulus intensity became progressively weaker or absent entirely – but would participants still report feeling it?

Schematic of the task, showing A) a visual checkerboard cue was paired with a mild electrical stimulus delivered to the ankle – participants then reported whether they felt the stimulus and their certainty; B) the distribution of trial types shifted over the course of the task - stimulus-absent trials became increasingly common

110

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

TLDR: Anxiety was associated with stronger, and possibly more persistent, prior beliefs about bodily sensations, which generated higher rates of percepts even when the sensory input was low or absent. For more detail, read on!

110

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

One explanation for somatic symptoms is that patients develop strong prior expectations of discomfort that outweigh the sensory information and dominate their perception. To test this, we built a new task that used Pavlovian conditioning to probe how tactile perceptual beliefs are acquired.

A depiction of bayesian inference showing how the prior belief and evidence combine to generate a distribution over posterior beliefs

131

Hugo Fleming @hugofleming.bsky.social · 19/08/2026

New paper alert! 📣📣 In this work, we investigated the origins of somatic symptoms – a common and debilitating feature of anxiety and depression. 'Anxiety is associated with the persistence of conditioned somatic sensations over time'

osf.io

OSF

2239

Hugo Fleming @hugofleming.bsky.social · 23/02/2026

Oh cool! Thanks Vanessa, this is a really thoughtful piece, and the framing around immuno-metabolic depression is exactly the kind of context we were hoping the work would connect to. The longitudinal CGM point is something that would be really exciting to see as well!

110

Hugo Fleming @hugofleming.bsky.social · 23/02/2026

Really pleased to see this commentary from @glassybrain.bsky.social on our paper. She situates our findings within the immunometabolic depression framework and points to some cool directions for future work, including longitudinal designs and measuring insulin resistance more directly. Take a look!

081

Hugo Fleming @hugofleming.bsky.social · 29/01/2026

Finally, if you found this interesting, follow me @hugofleming.bsky.social for more like this, and please share this thread so others can see it! bsky.app/profile/hugo...

030

Hugo Fleming @hugofleming.bsky.social · 29/01/2026

For the full story, you can find the paper out now in Biological Psychology! www.sciencedirect.com/science/arti.... Huge thanks to my brilliant co-authors: @annalisewhines.bsky.social Poppy Whelehan, @isabellau.bsky.social Katie Gallacher @saramehrhof.bsky.social and @camillanord.bsky.social

sciencedirect.com

151

Hugo Fleming @hugofleming.bsky.social · 29/01/2026

We discuss several possible explanations for this dissociation in the paper. Why does this matter? Because research into metabolic interoception in health and disease rests on understanding how metabolic expectations actually shape behaviour!

120

Hugo Fleming @hugofleming.bsky.social · 29/01/2026

Our results surprised us: people liked the caloric drink significantly more, but showed no PIT effect – no invigoration of behaviour from the CS+. This was strange given previous studies have implicated dopamine signalling in metabolic learning!

140

Hugo Fleming @hugofleming.bsky.social · 29/01/2026

53 participants drank two matched beverages daily for a week – one containing maltodextrin (calories), one without. Then they rated the drinks, and did the PIT task (a go/no-go with the flavoured drinks delivered by a mouthpiece). We verified people couldn’t consciously detect the maltodextrin.

140

Hugo Fleming @hugofleming.bsky.social · 29/01/2026

So we investigated whether metabolic learning actually drives differences in *human behaviour*. To do this, we made use of a phenomenon called Pavlovian-instrumental transfer (PIT), where a Pavlovian stimulus acquires incentive salience, biasing action.

140

Hugo Fleming @hugofleming.bsky.social · 29/01/2026

For animals to be motivated by a particular food, they first have to learn that it’s rewarding (containing calories, or other nutrients). In animals, metabolic learning is assessed through behaviour, but in humans it's usually measured with self-report questions – leading to mixed results.

130

Hugo Fleming @hugofleming.bsky.social · 29/01/2026

How does metabolic learning shape human behaviour? In our recent study www.sciencedirect.com/science/arti..., we found that it shapes flavour preferences but, surprisingly, not action. Thread 🧵

22313

Hugo Fleming @hugofleming.bsky.social · 12/10/2023

Fetching water is one thing, but it was when you had the servants bell installed in your office that I think it crossed a line

030

Reposted by Hugo Fleming

Laurence Hunt @lhuntneuro.bsky.social · 09/10/2023

Important/interesting-looking paper for those in computational psychiatry from Stefano Palminteri and colleagues psyarxiv.com/3u4gp/ #neuroscience #psychscisky

psyarxiv.com

32113

Hugo Fleming @hugofleming.bsky.social · 05/10/2023

Very true, and feels like a lesson you have to keep learning at every stage. ‘The better you want to be, the better the opportunities you have to say no to’

110

Reposted by Hugo Fleming

Jesus Ramirez- Bermudez @jrbneuropsiq.bsky.social · 03/10/2023

On #depression & allostatic overload 👀 In a sample with more than 200,000 individuals, severe depression was associated with the incidence of 29 conditions requiring hospital treatment (5-year follow-up): sleep disorders (HR, 5.97), diabetes (HR 5.15), ischemic heart disease (HR, 1.76), & others

198

Reposted by Hugo Fleming

Dr Lucy Foulkes @lucyfoulkes.bsky.social · 29/09/2023

New UCL study showing higher rates of MH symptoms in uni students vs non-students BUT - as the authors make clear in the study but the Guardian article doesn't ... www.theguardian.com/education/20...

theguardian.com

University students more at risk of depression than non-students – study

Higher levels of poor mental health in England could be linked to worries about money and academic achievement, findings suggest

144

Hugo Fleming @hugofleming.bsky.social · 28/09/2023

General career Q: what are the key skills one should learn at each level in academia? Or to put it another way, what are the skill bottlenecks you need to resolve to continue to progress?

152