Shan Siddiqi

Thanks for pointing this out! It's not a typo, the point was that they used incorrect statistics that yield overly strong p-values, and when we used valid stats we got a much weaker p-value.

A recent @natneuro.nature.com paper analyzed lesion network mapping and raised concerns about the validity of the method.
See below 👇 for our response.

www.biorxiv.org/content/10.6...

The methodological foundations of lesion network mapping remain sound https://www.biorxiv.org/content/10.64898/2026.02.24.707529v1

A recent paper in @natureportfolio.nature.com raised concerns about the lesion network mapping method. Our team of 16 coauthors analyzed >1000 lesions and 34 symptoms and found that "The methodological foundations of lesion network mapping remain sound" www.biorxiv.org/content/10.6...

To avoid becoming that person, I think it's important to follow all the standard things to take care of your brain - good sleep, exercise, etc. And when that's not working, have a low threshold for seeking help. Nobody was ever successful without a lot of help.

Usually when I've seen people cross that line, much of it is related to severe personal stressors (e.g. divorce, family illness, etc) on top of an untreated mental illness. It's sometimes helpful to keep that in mind when working with Grumpy.

It's hard to know when one is crossing the line from healthy scientific skepticism into jaded cynicism. Grumpy often doesn't realize they've crossed that line.

haha I'm not shy about telling people that you (and your 2014 Annals paper) introduced me to this field

We've long been aware of the challenges they describe... they're not fundamental flaws in the method, they're errors made by the authors. These errors are easy to make if you're not careful, and this is the standard suite of skills we teach our local trainees.

Yes, we'll see where it goes. Personally I think it should be retracted since the entire paper is predicated on the notion that they followed standard LNM methods, which is not true.

Thanks for sharing!
Tons of methodological flaws in the paper, but the fundamental problem imo is that it's a straw man argument that invokes a false equivalence fallacy. Of course LNM maps are similar, but they're also different. Just like I'm similar to a monkey, but also different.

btw the reason we don't use WM pathways is bc most lesions aren't monosynaptically connected to each other, so we wouldn't get data for every patient at every voxel, thus reducing statistical power. We've tried it before and it gives similar results but weaker, as expected (Joutsa 2022 Nat Med)

In this paper they created "synthetic" lesions using parcellations of resting-state networks, and then showed that they look like resting-state networks. Big whoop. They had access to real lesions and didn't use them?? Seems more likely that they tried it with real lesions and didn't report it.

Real lesions follow vascular distributions (stroke), white matter (MS), or near frontal bone (TBI). It's better to permute real lesion locations against other patients' clinical outcomes to get randomness.

Also the reason we don't compare to truly random outcomes is because that's an overly liberal method. Comparing to random iterations of real outcomes - what we actually do - is more conservative. In one study we compared to random outcomes bc a reviewer asked for it (Makhlouf 2024 Nat Men Hlth).

They claim to follow our published methods, but they didn't actually do that. I re-ran some of their analyses using our published permutation-based methods for spatial correspondence, and suddenly p<0.001 becomes p=0.36.

I've reported this to the journal and will expect a retraction.

The analyses of "simulated" lesions are invalid because their "simulated" lesions were actually parcels that were derived from the same connectome, and are thus more likely to just yield the task-positive and task-negative network.

We don't get this result when using real lesions.

They claim to follow our published methods, but they didn't actually do that. I re-ran some of their analyses using our published permutation-based methods for spatial correspondence, and suddenly p<0.001 becomes p=0.36.

I've reported this to the journal and will expect a retraction.

This will be an easy paper to refute. The central argument is "things that are similar can't also be different." And similarity was measured using a metric that yields overly inflated measures of similarity, which we've explicitly recommended not using.

Don't worry, the methods they used are known to yield inflated metrics of spatial correspondence.

I just rechecked the correspondence between two of my papers that they report to be p<0.001, and actually it's p=0.36 if you follow our recommended permutation methods (actually based on your methods).

DLPFC Stimulation Suppresses High-Frequency Neural Activity in the Human sgACC Transcranial magnetic stimulation (TMS) to the dorsolateral prefrontal cortex (DLPFC) is hypothesized to relieve symptoms of depression by inhibiting activity in the subgenual anterior cingulate corte...

When we use TMS to treat depression, we think it alters neural activity in a deep brain structure known as the subgenual anterior cingulate (sgACC). But this is hard to test directly, until our preprint with @coreykeller.bsky.social, @neuro-engineer.bsky.social, Nick Trapp, and Aaron Boes (UIowa) 1/

A generalized epilepsy network derived from brain abnormalities and deep brain stimulation - Nature Communications Ji et al. identify an idiopathic generalised epilepsy network that links heterogeneously distributed brain abnormalities to a common brain network and deep brain stimulation sites which reduce general...

Is generalized epilepsy a seizure of the whole brain or a specific brain network?

In our new paper out now in @naturecomms.bsky.social, we investigate this by combining brain abnormalities and DBS with the human connectome.

Paper: nature.com/articles/s41...

A 🧵 below:

Out now in Brain

A distinct pattern of brain damage intensifies political involvement, regardless of whether you're liberal, conservative, Democrat, or Republican.

With coauthors S.Balters,S. Cohen-Zimerman,G.Zamboni,J.Grafman

academic.oup.com/brain/advanc...

@braincircuits.bsky.social

Looking toward to our ANPA symposium on interventional neuropsychiatry - kicked off w a brilliant talk & overview on TMS by Nick Trapp.

Just to clarify, training won't be a prerequisite to providing brain stimulation... just an optional pathway to learn advanced techniques

If this hasn't been reported before, I call dibs on the eponym "Shan's postulate"

Am I missing something? Is this just a mathematical fact? Or does it need to be validated empirically? It seems like most psychiatric biomarker studies miss this point, as does the culture around "target engagement" in the experimental therapeutics pipeline.

If I'm correct, and if the depression scale has test-retest reliability of z(r) = 0.8ish and fMRI biomarker has test-retest reliability of z(r) = 0.6ish, then no fMRI biomarker model can accurately explain variance in depression better than r = 0.5.

All biomarker studies should assess reliability.

I think the theoretical limit of explainable variance in a model should be less than the product of the test-retest reliabilities of the input variables. If the explained variance in a study exceeds that value (as it often does in the psychiatric literature), the model must be overfit.

Out now in Brain Stimulation

Towards accredited training in brain stimulation: proceedings from the Brain Stimulation Subspecialty Summits (BraSSS)

with co-host Nolan Williams and 23 other awesome co-authors including some from @braincircuits.bsky.social

www.brainstimjrnl.com/article/S193...