Manjari Narayan | @Neurostats
@neurostats.org
AI in Bio & Health & Therapeutic Development
Bio: https://linktr.ee/mnarayan
Substack: https://blog.neurostats.org
Peek into my brain: notes.manjarinarayan.org
Previously @dynotx @StanfordMed PhD@RiceU_ECE | BS@ECEILLINOIS
🧪🧮⚕️🧬🧠🖥🤖📈✍️🩺👩📈📉
Bio: https://linktr.ee/mnarayan
Substack: https://blog.neurostats.org
Peek into my brain: notes.manjarinarayan.org
Previously @dynotx @StanfordMed PhD@RiceU_ECE | BS@ECEILLINOIS
🧪🧮⚕️🧬🧠🖥🤖📈✍️🩺👩📈📉
Pinned
0/
Folks are asking for resources for learning about causal inference.
Here is a comprehensive reading list since no one address all issues —
What is causality?
When is a causal relationship is knowable?
How to infer causal relationships given from noisy and limited data?
Folks are asking for resources for learning about causal inference.
Here is a comprehensive reading list since no one address all issues —
What is causality?
When is a causal relationship is knowable?
How to infer causal relationships given from noisy and limited data?
Reposted by Manjari Narayan | @Neurostats
I think the problem here is that most people who do epidemiological research - including many who might call themselves 'epidemiologists' - have not actually had any epidemiology training. I wish it was a protected title. With an exam that would require proving you understand residual confounding!
July 17, 2025 at 2:05 PM
I think the problem here is that most people who do epidemiological research - including many who might call themselves 'epidemiologists' - have not actually had any epidemiology training. I wish it was a protected title. With an exam that would require proving you understand residual confounding!
Reposted by Manjari Narayan | @Neurostats
For example, current evidence is that LLMs rely too much on interpolation rather than discovering underlying invariant causal abstractions.
June 26, 2025 at 5:36 PM
For example, current evidence is that LLMs rely too much on interpolation rather than discovering underlying invariant causal abstractions.
This appears to be a very problematic weaponization of casual reasoning that feels 'rigorous' because it involves a selective presentation of evidence. Some very excellent researchers have looked into the problems from air pollution.
#CausalSky #MetaSky
www.theguardian.com/technology/n...
#CausalSky #MetaSky
www.theguardian.com/technology/n...
Inside a plan to use AI to amplify doubts about the dangers of pollutants
Risk analyst Tony Cox’s work has been backed by the chemical lobby, and some health experts are alarmed
www.theguardian.com
June 27, 2025 at 3:15 PM
This appears to be a very problematic weaponization of casual reasoning that feels 'rigorous' because it involves a selective presentation of evidence. Some very excellent researchers have looked into the problems from air pollution.
#CausalSky #MetaSky
www.theguardian.com/technology/n...
#CausalSky #MetaSky
www.theguardian.com/technology/n...
Exactly
> The people who care about a topic enough to research it will often have ideological motivations. The ideological influence is there. That doesn’t mean it’s bad science.
> The people who care about a topic enough to research it will often have ideological motivations. The ideological influence is there. That doesn’t mean it’s bad science.
NIH plan to remove ideological influence from science. How does this fit in the junk science being promoted by the U.S. Dept of Health and Human Services?
statmodeling.stat.columbia.edu/2025/06/20/n...
statmodeling.stat.columbia.edu/2025/06/20/n...
NIH plan to remove ideological influence from science. How does this fit in the junk science being promoted by the U.S. Dept of Health and Human Services? | Statistical Modeling, Causal Inference, a...
statmodeling.stat.columbia.edu
June 22, 2025 at 8:48 PM
Exactly
> The people who care about a topic enough to research it will often have ideological motivations. The ideological influence is there. That doesn’t mean it’s bad science.
> The people who care about a topic enough to research it will often have ideological motivations. The ideological influence is there. That doesn’t mean it’s bad science.
Every few years there is a new article like this, but not much will change unless scientists stop relying on GraphPad Prism that doesn't offer any of this and have something they can use more easily.
LLM powered workflows might change this going forward.
LLM powered workflows might change this going forward.
A brief guide to statistical analysis of grouped data in preclinical research www.nature.com/articles/s42... #datascience #bioinformatics #statistics #mlsky
A brief guide to statistical analysis of grouped data in preclinical research - Nature Metabolism
Clustering and nesting (C&N) arise in many preclinical studies, such as when animals are group-housed or share litters, or in cell culture. Ignoring C&N undermines the validity of analyses. He...
www.nature.com
June 22, 2025 at 5:51 PM
Every few years there is a new article like this, but not much will change unless scientists stop relying on GraphPad Prism that doesn't offer any of this and have something they can use more easily.
LLM powered workflows might change this going forward.
LLM powered workflows might change this going forward.
Unless you are using inferential clustering methods that are designed to severely test any hypotheses against the baseline that "how often would you find clusters if the data was generated by these high dimensional null patterns or if data were corrupted by structured measurement error"
1) LCAs are designed to find classes. So when a paper's Aim I is "We'll use LCA to see if there are classes", it's like saying "We're going to estimate a mean and see if there's an SD". It's that dumb
June 21, 2025 at 1:24 PM
Unless you are using inferential clustering methods that are designed to severely test any hypotheses against the baseline that "how often would you find clusters if the data was generated by these high dimensional null patterns or if data were corrupted by structured measurement error"
Can you demonstrate the extent and degree of different kinds of sequential selection biases from a #CausalSky #StatSky #MedSky perspective in public clinical trial datasets?
DM me if you are interested.
For instance in TrialBench?
github.com/ML2Health/ML...
DM me if you are interested.
For instance in TrialBench?
github.com/ML2Health/ML...
ML2ClinicalTrials/Trialbench at main · ML2Health/ML2ClinicalTrials
Contribute to ML2Health/ML2ClinicalTrials development by creating an account on GitHub.
github.com
June 21, 2025 at 1:20 PM
Can you demonstrate the extent and degree of different kinds of sequential selection biases from a #CausalSky #StatSky #MedSky perspective in public clinical trial datasets?
DM me if you are interested.
For instance in TrialBench?
github.com/ML2Health/ML...
DM me if you are interested.
For instance in TrialBench?
github.com/ML2Health/ML...
Reposted by Manjari Narayan | @Neurostats
She wrote this in May…
Can’t take it with you
I am in hospice care and reflecting a lot on what a good life is.
open.substack.com
June 20, 2025 at 9:09 PM
She wrote this in May…
Reposted by Manjari Narayan | @Neurostats
the Mesoscopic Integrated Neuroimaging Data (MIND) platform will soon be opening also for outside researchers leveraging our new 15.2 T 🐁 MRI, SHIELD clearing and fully upgraded lightsheet microscope plus computational resources developed by @neuroak.bsky.social to handle large datasets.
Great news just announced. @neuroak.bsky.social is leading a new funded @braincanada.bsky.social platform that promises to revolutionize the study of neurodegeneration by tackling Patho-connectomics Merging MRI and lightsheet analysis in new models of disease www.schulich.uwo.ca/about/news-e...
Researchers launch brain imaging platform with $2.9M boost from Brain Canada - Schulich School of Medicine & Dentistry - Western University
A new high-resolution imaging platform is revealing the brain in extraordinary detail – and sharing the data with scientists worldwide.
www.schulich.uwo.ca
June 20, 2025 at 3:01 PM
the Mesoscopic Integrated Neuroimaging Data (MIND) platform will soon be opening also for outside researchers leveraging our new 15.2 T 🐁 MRI, SHIELD clearing and fully upgraded lightsheet microscope plus computational resources developed by @neuroak.bsky.social to handle large datasets.
Neurobullshit is alive and well in education research. Reminds me of the good old days of @neuroskeptic.bsky.social on this 10+ years ago
I think the conclusion of this study is likely to be valid however it is worth noting that any attempts to localise cognitive tasks such as writing to a specific brain activation/connectivity is problematic because that's not how the brain works.
1/
1/
June 19, 2025 at 9:21 PM
Neurobullshit is alive and well in education research. Reminds me of the good old days of @neuroskeptic.bsky.social on this 10+ years ago
Also, longitudinal studies don't fix causality problems. Fortunately, they help you discover entirely new kinds of confounding.
To re-up this, this also applies to latent growth curve models 😭 😭 😭
No, I'm afraid you cannot solve a fundamental identification problem by applying a latent growth curve model.
No, I'm afraid you cannot solve a fundamental identification problem by applying a latent growth curve model.
PSA: Don’t trust anyone who tells you that you can identify age or period or cohort effects simply by applying the right statistical model to the right type of data. This is fundamentally misunderstanding the nature of the age-period-cohort problem!>
June 18, 2025 at 2:17 PM
Also, longitudinal studies don't fix causality problems. Fortunately, they help you discover entirely new kinds of confounding.
Reposted by Manjari Narayan | @Neurostats
If you use a model dialogically, it’s effectively guiding you through the free-writing/revision cycle composition teachers recommend.
I get why this mode of use isn’t very visible. All the rhetoric around AI has been that it’s “generative” and does the work for you. Things like the “Dear Sydney” +
I get why this mode of use isn’t very visible. All the rhetoric around AI has been that it’s “generative” and does the work for you. Things like the “Dear Sydney” +
June 18, 2025 at 10:52 AM
If you use a model dialogically, it’s effectively guiding you through the free-writing/revision cycle composition teachers recommend.
I get why this mode of use isn’t very visible. All the rhetoric around AI has been that it’s “generative” and does the work for you. Things like the “Dear Sydney” +
I get why this mode of use isn’t very visible. All the rhetoric around AI has been that it’s “generative” and does the work for you. Things like the “Dear Sydney” +
5 years later, it is finally on youtube. Thanks @ohbmofficial.bsky.social
youtu.be/X5syQN9ZkQE?...
#neuroskyence #CausalSky
youtu.be/X5syQN9ZkQE?...
#neuroskyence #CausalSky
June 17, 2025 at 9:07 PM
5 years later, it is finally on youtube. Thanks @ohbmofficial.bsky.social
youtu.be/X5syQN9ZkQE?...
#neuroskyence #CausalSky
youtu.be/X5syQN9ZkQE?...
#neuroskyence #CausalSky
👀 A performative prediction paper
Group member Philippe van Basshuysen's paper "When Predictions are More Than Predictions: Self-Fulfilling Performativity and the Road Towards Morally Responsible Predictive Systems" has just been accepted for publication
www.academia.edu/129990683/Wh...
www.academia.edu/129990683/Wh...
When Predictions are More Than Predictions: Self-Fulfilling Performativity and the Road Towards Morally Responsible Predictive Systems
Some predictive systems do not merely predict, but their predictions shape and steer the world towards certain outcomes rather than others; they are performative. When predictive systems are performat...
www.academia.edu
June 17, 2025 at 4:49 PM
👀 A performative prediction paper
Reposted by Manjari Narayan | @Neurostats
What’s common knowledge in your field but shocks outsiders?
The vast majority of archival documents are neither digitized nor searchable online. 🗃️
The vast majority of archival documents are neither digitized nor searchable online. 🗃️
What’s common knowledge in your field but shocks outsiders?
Not everything you think should be said out loud.
Not everything you think should be said out loud.
June 17, 2025 at 4:15 PM
What’s common knowledge in your field but shocks outsiders?
The vast majority of archival documents are neither digitized nor searchable online. 🗃️
The vast majority of archival documents are neither digitized nor searchable online. 🗃️
Reductionism never falls out of fashion because one has to embrace theory to question it.
That's partly why I wrote How Life Works. This idea that if we understand the molecules/genes then we understand the biology is corrosive - evidently even at the highest levels.
June 17, 2025 at 4:41 PM
Reductionism never falls out of fashion because one has to embrace theory to question it.
FDR metrics only offer error-control of the false discovery proportion under expectation. It is already a liberal standard of evidence when you realize that. Most don't.
If one wants to defend a paper in nature, you might find AI reasoning agents quite useful to check your work.
#StatSky 🖥️ 🧬 📈💻
If one wants to defend a paper in nature, you might find AI reasoning agents quite useful to check your work.
#StatSky 🖥️ 🧬 📈💻
June 17, 2025 at 1:08 AM
FDR metrics only offer error-control of the false discovery proportion under expectation. It is already a liberal standard of evidence when you realize that. Most don't.
If one wants to defend a paper in nature, you might find AI reasoning agents quite useful to check your work.
#StatSky 🖥️ 🧬 📈💻
If one wants to defend a paper in nature, you might find AI reasoning agents quite useful to check your work.
#StatSky 🖥️ 🧬 📈💻
Reposted by Manjari Narayan | @Neurostats
What was interesting to me was that, despite my best efforts, SCB was unable to understand that this was a framing of the observations, and not simply a description of them.
The response was that to be curious about this framing revealed total ignorance of the field, despite the following...
/3
The response was that to be curious about this framing revealed total ignorance of the field, despite the following...
/3
May 24, 2025 at 7:06 PM
What was interesting to me was that, despite my best efforts, SCB was unable to understand that this was a framing of the observations, and not simply a description of them.
The response was that to be curious about this framing revealed total ignorance of the field, despite the following...
/3
The response was that to be curious about this framing revealed total ignorance of the field, despite the following...
/3
Awesome answers. Can we have a follow-up meta science workshop?
Serious question:
why is theory development in neuroscience so damn hard?
why is theory development in neuroscience so damn hard?
June 16, 2025 at 2:59 PM
Awesome answers. Can we have a follow-up meta science workshop?
The generous variation is that the assumption is mechanisms are separable, can be studied independently and then combined together in a way that will converge to the truth.
Brace, brace: neuro-centrism, cos the brain is treated like an isolated solution, divorced from body and behaviour
June 16, 2025 at 2:58 PM
The generous variation is that the assumption is mechanisms are separable, can be studied independently and then combined together in a way that will converge to the truth.
Reposted by Manjari Narayan | @Neurostats
Apart from at a few over-plucked trees, for many subtopics it is not clear how to make a good theoretical contribution to the area. Involves inventing new plucking tools, deciding which fruit are both within grasp and edible for your experimental colleagues
June 7, 2025 at 2:36 PM
Apart from at a few over-plucked trees, for many subtopics it is not clear how to make a good theoretical contribution to the area. Involves inventing new plucking tools, deciding which fruit are both within grasp and edible for your experimental colleagues
This is very interesting. It is indeed problematic that psychotherapy efficaciousness is not being evaluated. But I'm not sure about the anesthetic analogy. The problem is that the two together represent a kind of non-additive therapeutic interaction in a way that isn't necessary with anesthesia.
With Natalie Gukasyan, @leorroseman.bsky.social & @pliknaitzky.bsky.social
LINK to the article: journals.sagepub.com/doi/10.1177/...
LINK to the article: journals.sagepub.com/doi/10.1177/...
Reframing psychedelic regulation: Tools, not treatments - Max Wolff, Natalie Gukasyan, Leor Roseman, Paul Liknaitzky, 2025
Current regulation frameworks for medicines struggle to address the combination of pharmacological and psychotherapeutic elements in psychedelic therapy. We pro...
journals.sagepub.com
June 16, 2025 at 12:19 PM
This is very interesting. It is indeed problematic that psychotherapy efficaciousness is not being evaluated. But I'm not sure about the anesthetic analogy. The problem is that the two together represent a kind of non-additive therapeutic interaction in a way that isn't necessary with anesthesia.
Reposted by Manjari Narayan | @Neurostats
I wish some of the folks with terminated grants- especially Harvard, Columbia folks - would post the summary statements from NIH study sections about their grants. The public should know what independent scientists thought about the projects that are being terminated. #PublicHealth #EpiSky #EndAlz
June 15, 2025 at 8:07 PM
I wish some of the folks with terminated grants- especially Harvard, Columbia folks - would post the summary statements from NIH study sections about their grants. The public should know what independent scientists thought about the projects that are being terminated. #PublicHealth #EpiSky #EndAlz
Reposted by Manjari Narayan | @Neurostats
One reason I hate the term "Polynomial Regression" so much is because linear regression means "linear in parameters", not "linear in variables". I'm not calling it logarithmic regression either when logging some variables. Still linear regression. Stop making squaring a variable sound like magic.
June 15, 2025 at 4:23 PM
One reason I hate the term "Polynomial Regression" so much is because linear regression means "linear in parameters", not "linear in variables". I'm not calling it logarithmic regression either when logging some variables. Still linear regression. Stop making squaring a variable sound like magic.
Reposted by Manjari Narayan | @Neurostats
Not a single AI science agent paper I've read so far tackles the issue that actual scientific discovery bears no resemblance to the hypothetico-deductive veneer given by the papers describing results.
So asking agents to match the literature and/or solve problems in this way feels very off ...
So asking agents to match the literature and/or solve problems in this way feels very off ...
January 11, 2025 at 12:22 AM
Not a single AI science agent paper I've read so far tackles the issue that actual scientific discovery bears no resemblance to the hypothetico-deductive veneer given by the papers describing results.
So asking agents to match the literature and/or solve problems in this way feels very off ...
So asking agents to match the literature and/or solve problems in this way feels very off ...