For years, we've known that running a standard t-test on cross-validation folds violates sample independence. We wanted to see how widespread this issue actually is. The result? 97% of the studies used an invalid statistical test. 🧵👇

Thomas Yeo@bttyeo.bsky.social · 3mo ago

In a meta-analysis of 210 biomedical AI studies that statistically compared models under cross-validation, 97% used invalid statistical tests. Here's our new preprint doi.org/10.64898/202... led by @tianchu.bsky.social @hetuli.bsky.social @shaoshiz.bsky.social @nichols.bsky.social 1/N

@nichols.bsky.social collaborated with researchers at the National University of Singapore on a recent study published in @nature.com on how longer duration fMRI brain scans reduce costs and improve prediction accuracy for AI models. Read more about the study below 👇

Thomas Yeo@bttyeo.bsky.social · last yr.

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

Just dropped in @natcomms.nature.com: we show that re-engaging a thalamic–ventral tegmental circuit with deep brain stimulation can reignite consciousness in patients with severe brain injury. Work led by Aaron Warren, with @andreashorn.org @foxmdphd.bsky.social @ others! tinyurl.com/4kz8j89b

A human brain network linked to restoration of consciousness after deep brain stimulation - Nature Communications

In people with severe brain injuries, stimulation restored consciousness by engaging a deep brain circuit for wakefulness—revealing a target that may also guide treatment in stroke and epilepsy.

nature.com

What a fantastic effort. Truly inspiring to see brilliant people dig deeply into these meta scientific issues. This is the best time to be doing neuroimaging.

Thomas Yeo@bttyeo.bsky.social · last yr.

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

I'm so proud to see this great paper finally published in @nature.com!

Thomas Yeo@bttyeo.bsky.social · last yr.

9/ Check out the actual study for many more analyses, e.g., phenotypic variation, scan parameters, signal to noise ratio, etc! doi.org/10.1038/s415... Thank you to the editor @meharpist.bsky.social and anonymous reviewers for the many helpful suggestions, which greatly improved the study.

Our Nature paper on the hashtag#scaling hashtag#behavior and economics of hashtag#machine hashtag#learning predictions in high-dimensional brain scans is out ! Congrats to the whole team. www.nature.com/articles/s41...

Really nice study, and extends some of the ideas developed in this paper pubmed.ncbi.nlm.nih.gov/32673043/

Thomas Yeo@bttyeo.bsky.social · last yr.

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

A super important and well designed study. Curious if those who took such interest in the original "BWAS needs impossibly huge n" will pay any attention to it

Thomas Yeo@bttyeo.bsky.social · last yr.

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

This new Yeo Lab tool should immediately and permanently replace sample-size-only power calculations for functional MRI. www.nature.com/articles/s41...

Longer scans boost prediction and cut costs in brain-wide association studies - Nature

Although the number of participants is important for phenotypic prediction accuracy in brain-wide association studies using functional MRI, scanning for at least 30 min offers the greatest cost effect...

nature.com

Thomas Yeo@bttyeo.bsky.social · last yr.

8/ In contrast to standard power calculations, our results suggest that jointly optimizing sample size and scan time can boost prediction accuracy while cutting costs. For more complex study design, you can check out our calculator: thomasyeolab.github.io/OptimalScanT...

Just incredible results from a massive effort— moves the field forward. Bravo!!!

Thomas Yeo@bttyeo.bsky.social · last yr.

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

Nature research paper: Longer scans boost prediction and cut costs in brain-wide association studies go.nature.com/3IME4aA

Longer scans boost prediction and cut costs in brain-wide association studies - Nature

Although the number of participants is important for phenotypic prediction accuracy in brain-wide association studies using functional MRI, scanning for at least 30 min offers the greatest cost effectiveness.

go.nature.com

Big congrats to @bttyeo.bsky.social and team on this impressive and important work!

Thomas Yeo@bttyeo.bsky.social · last yr.

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

For me, this work is a classic @ohbmofficial.bsky.social story: In 2023 I wasn't working with @bttyeo.bsky.social but I overheard him at his poster pointing to some accuracy curves saying "I don't why they have this particular shape". That kicked off the collab that led to these results.

Thomas Yeo@bttyeo.bsky.social · last yr.

3/11 Tom's model explains empirical prediction accuracies well across 76 phenotypes from 9 resting-fMRI & task-fMRI datasets (R2 = 0.89), spanning many scanners, acquisitions, racial groups, disorders & ages. Does this mean that we should collect large datasets with short scans?

Everyone should try out the Trandiagnostic Connectome Project (TCP) dataset! Openly available on @openneuro.bsky.social

Sidhant Chopra@sidchop.bsky.social · last yr.

V useful paper by @bttyeo.bsky.social @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social in @nature.com. Scan longer if you want to predict behav using fMRI and save $. Great use of the TCP data: (pmc.ncbi.nlm.nih.gov/articles/PMC...).

V useful paper by @bttyeo.bsky.social @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social in @nature.com. Scan longer if you want to predict behav using fMRI and save $. Great use of the TCP data: (pmc.ncbi.nlm.nih.gov/articles/PMC...).

Thomas Yeo@bttyeo.bsky.social · last yr.

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

Super thankful to @bttyeo.bsky.social @csabaorban.bsky.social and @shaoshiz.bsky.social for pouring in all the effort to make this work possible!

Thomas Yeo@bttyeo.bsky.social · last yr.

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

🚨Thrilled to share our latest work just published in @nature.com where we looked into the optimal fMRI scan time for brain-wide association studies (BWAS) 🧠⏱️! Full thread below👇:

Thomas Yeo@bttyeo.bsky.social · last yr.

1/11 Excited to share our @Naturestudy led by @leonooi.bsky.social @csabaorban.bsky.social @shaoshiz.bsky.social AI performance is known to scale with logarithm of sample size (Kaplan 2020), but in many domains, sample size can be # participants or # measurements... doi.org/10.1038/s415...

How does the human brain coordinate hierarchical cortical development? Our work in Nature Neuroscience identifies a role for thalamocortical structural connectivity in the expression of hierarchical periods of cortical plasticity & environmental receptivity in youth 🧵 www.nature.com/articles/s41...

Check out our latest open data release. n=240, most with a dsm-5 dx with extensive phenotying (~100 scales/subscale), rest and task functional imaging. See @carrisacocuzza.bsky.social's thread below for deets and links 👇🏾👇🏾👇🏾

Carrisa V. Cocuzza, PhD (she/her)@carrisacocuzza.bsky.social · last yr.

🚨 Dataset & Manuscript alert! 🚨 The Transdiagnostic Connectome Project (TCP) manuscript is now available @natureportfolio.nature.com Scientific Data! 🎉 www.nature.com/articles/s41... 🧵1👇

I love the work, not only because it speed up FIC models a lot, but also how it saves poor students from grad student descent 🤣🤣

Thomas Yeo@bttyeo.bsky.social · last yr.

While the world burns, we cook up a new preprint! doi.org/10.1101/2025... Biophysical modeling is a key tool to derive mechanistic insights into the brain. These models are governed by biologically meaningful parameters (unlike artificial neural networks), but the dirty secret ... 1/N

Can deep learning help us solve dynamical systems problems, particularly those used in neural mass models? Check out this preprint to read about the perks...

Thomas Yeo@bttyeo.bsky.social · last yr.

While the world burns, we cook up a new preprint! doi.org/10.1101/2025... Biophysical modeling is a key tool to derive mechanistic insights into the brain. These models are governed by biologically meaningful parameters (unlike artificial neural networks), but the dirty secret ... 1/N

Check our latest preprint led by the amazing @tianchu.bsky.social and @tianfang.bsky.social where we speed up the tedious parameter optimization process for biophysical modelling

Thomas Yeo@bttyeo.bsky.social · last yr.

While the world burns, we cook up a new preprint! doi.org/10.1101/2025... Biophysical modeling is a key tool to derive mechanistic insights into the brain. These models are governed by biologically meaningful parameters (unlike artificial neural networks), but the dirty secret ... 1/N

🚨 Predicting Alzheimer's Progression 🚨 A thread 🧵 1/ Accurate prediction of Alzheimer’s progression is critical for early intervention. How can we make predictions more precise and generalizable? 🧠✨ 📝 Read the preprint led by @chen-zhang.bsky.social : doi.org/10.1101/2024...