Pinned Tweet
💊 ATTENTION ALL DOCTORS 🤖
If you haven’t become AI native yet, here is my plea to you.
Start now.
It might feel overwhelming.
It might feel unethical.
It might feel existential.
But it is not.
It is the greatest enabler we have ever experienced in our lifetime 🧵
If you haven’t become AI native yet, here is my plea to you.
Start now.
It might feel overwhelming.
It might feel unethical.
It might feel existential.
But it is not.
It is the greatest enabler we have ever experienced in our lifetime 🧵
2
0
13
1,737
Being at Pareto for one model across multiple design tasks and substrates is pretty stacked ngl
Avengers assemble: 1 year BakerLab + NVIDIA GenAIR = 🔥 rfd4-proteina.org
Presenting RFD4-Proteina, our latest protein design model, significantly better and faster than everything we compared against, both in silico and in the lab! 1/30
0
0
3
458
📝 Literature
I have a daily scraper for new papers, posts on X, or blogs on topics to learn about.
I have one for:
- AI x Bio
- Psych drug discovery
- Psych news
- Macroeconomics
- AI fundamentals
I have a daily feed that tells me one concept to go deep into, and then a news brief. That way I’m learning stuff to build my foundations and also what’s cutting edge. Super important for me on the AI side where I’m still new and learning. Same for macroecon.
How to set this up? Copy paste this and tell Grok Bot to do this for you.
I have a daily scraper for new papers, posts on X, or blogs on topics to learn about.
I have one for:
- AI x Bio
- Psych drug discovery
- Psych news
- Macroeconomics
- AI fundamentals
I have a daily feed that tells me one concept to go deep into, and then a news brief. That way I’m learning stuff to build my foundations and also what’s cutting edge. Super important for me on the AI side where I’m still new and learning. Same for macroecon.
How to set this up? Copy paste this and tell Grok Bot to do this for you.
1
0
0
227
I’m honestly so worried that the PRIMARY zeitgeist in medicine is one of skeptical resistance to being AI native.
Most faculty are still unclear about the different models, their strengths and weaknesses, and how to integrate AI effectively.
It’s now easier than ever to do so.
So my plea is this: learn how to use AI because it will make you the best clinician you can be. For yourself. For your peers. And for your patients.
Most faculty are still unclear about the different models, their strengths and weaknesses, and how to integrate AI effectively.
It’s now easier than ever to do so.
So my plea is this: learn how to use AI because it will make you the best clinician you can be. For yourself. For your peers. And for your patients.
3
0
0
196
🚨 [BREAKING]
Brain scans now show that for ~60% of patients taking sertraline, their brain network changes and clinical responses are purely PLACEBO-driven.
We might be giving SSRIs to 60% of people who don't need it 🧵
Brain scans now show that for ~60% of patients taking sertraline, their brain network changes and clinical responses are purely PLACEBO-driven.
We might be giving SSRIs to 60% of people who don't need it 🧵
5
10
55
4,842
🔎 The replication cohort lacked a placebo control
To prove these brain signatures weren't just a one-off, the authors validated them in the CAN-BIND-1 escitalopram cohort.
But because CAN-BIND-1 lacked a placebo arm, we cannot definitively confirm if escitalopram induces the cebo-to-drug responders as sertraline.
We also only have 1-week and 2-week snapshots. We do not know if the 60% of patients who looked like placebo responders at week one eventually developed the true drug-specific brain changes by week four.
Finally, the study failed to identify an early fMRI signature that reliably predicts continuous individual-level treatment response to sertraline.
To prove these brain signatures weren't just a one-off, the authors validated them in the CAN-BIND-1 escitalopram cohort.
But because CAN-BIND-1 lacked a placebo arm, we cannot definitively confirm if escitalopram induces the cebo-to-drug responders as sertraline.
We also only have 1-week and 2-week snapshots. We do not know if the 60% of patients who looked like placebo responders at week one eventually developed the true drug-specific brain changes by week four.
Finally, the study failed to identify an early fMRI signature that reliably predicts continuous individual-level treatment response to sertraline.
1
1
0
352
💡 Takeaways
The modest benefit SSRIs have over placebos in major depression trials isn't just because the drugs are inherently weak across the board.
It is because clinical trials are blending two entirely different biological subpopulations: a minority having a potent e, and a majority having a placebo response to an active pill.
This paper provides strong evidence that we can see that divergence in functional brain networks within seven days of the first dose.
If we can develop pre-treatment fMRI signatures to identify that 40% before prescribing, we stop relying on trial-and-error and move toward actual precision psychiatry.
But this harder than it sounds, because it seems like you need to "give" the treatment first and measure a week later to know. We still don't yet have a way to do a single scan and predict placebo response characteristics with any certainty.
Solving the challenge of the placebo effect will truly be field-changing for psychiatry.
Source: Tong et al. bioRxiv 2025 — doi.org/10.1101/2025.0…
The modest benefit SSRIs have over placebos in major depression trials isn't just because the drugs are inherently weak across the board.
It is because clinical trials are blending two entirely different biological subpopulations: a minority having a potent e, and a majority having a placebo response to an active pill.
This paper provides strong evidence that we can see that divergence in functional brain networks within seven days of the first dose.
If we can develop pre-treatment fMRI signatures to identify that 40% before prescribing, we stop relying on trial-and-error and move toward actual precision psychiatry.
But this harder than it sounds, because it seems like you need to "give" the treatment first and measure a week later to know. We still don't yet have a way to do a single scan and predict placebo response characteristics with any certainty.
Solving the challenge of the placebo effect will truly be field-changing for psychiatry.
Source: Tong et al. bioRxiv 2025 — doi.org/10.1101/2025.0…
1
2
5
344
🚨 [BREAKING] Schizophrenia genetics will never be the same.
The largest rare variant study of schizophrenia just dropped.
Tripling the sample size of v1.0, SCHEMA 2.0 identifies 16 genes at exome-wide significance and 40 at a 5% FDR.
The wildest part is how it directly proves Conenfy🧵
The largest rare variant study of schizophrenia just dropped.
Tripling the sample size of v1.0, SCHEMA 2.0 identifies 16 genes at exome-wide significance and 40 at a 5% FDR.
The wildest part is how it directly proves Conenfy🧵
11
42
212
36,083
🔎 Ascertainment shifts and the missing 1.0 genes
Interestingly, three exome-wide genes from SCHEMA 1.0 (GRIA3, GRIN2A, CACNA1G) lost their top-tier significance here.
These genes are strongly tied to severe developmental delay. The authors note newly added cohorts likely iatric-onset cases, diluting the signal.
Furthermore, despite the large odds ratios, estimates for individual genes are highly unstable with wide confidence intervals because carriers are so rare. Only 1.5% of cases carry a PTV in one of these 40 genes.
We cannot use this list as a diagnostic tool. The absence of a mutation tells you nothing about diagnosis, and presence is not deterministic (although the risk is increased, often dramatically).
Interestingly, three exome-wide genes from SCHEMA 1.0 (GRIA3, GRIN2A, CACNA1G) lost their top-tier significance here.
These genes are strongly tied to severe developmental delay. The authors note newly added cohorts likely iatric-onset cases, diluting the signal.
Furthermore, despite the large odds ratios, estimates for individual genes are highly unstable with wide confidence intervals because carriers are so rare. Only 1.5% of cases carry a PTV in one of these 40 genes.
We cannot use this list as a diagnostic tool. The absence of a mutation tells you nothing about diagnosis, and presence is not deterministic (although the risk is increased, often dramatically).
1
0
8
1,417
💡 Takeaways
SCHEMA 2.0 reinforces the notion that scaling up rare variant sequencing consistently yields highly penetrant biology.
We now have 40 high-to-very-high-confidence risk genes for schizophrenia, tightly linking the disorder to early neurodevelopmental pathways and ions.
The overlap of CHRM4 with the newest class of antipsychotics provides validation for a genetics-based target-ID approach using RVAS for neuropsychiatric disorders, showing it can find genuine, modifiable drug targets.
The open question is how we translate the other 39 genes into safe, targeted therapeutics for the remaining 98% of patients who do not carry these exact mutations.
These genes are often difficult to target (chromatin modifiers, ubiquitin ligases); but that's changing with the increasing ease of binder design with all kinds of properties that make the "undruggable" druggable.
Exciting times to be alive in Psychiatry.
Source: Sealock et al. bioRxiv 2026 — doi.org/10.64898/2026.…
SCHEMA 2.0 reinforces the notion that scaling up rare variant sequencing consistently yields highly penetrant biology.
We now have 40 high-to-very-high-confidence risk genes for schizophrenia, tightly linking the disorder to early neurodevelopmental pathways and ions.
The overlap of CHRM4 with the newest class of antipsychotics provides validation for a genetics-based target-ID approach using RVAS for neuropsychiatric disorders, showing it can find genuine, modifiable drug targets.
The open question is how we translate the other 39 genes into safe, targeted therapeutics for the remaining 98% of patients who do not carry these exact mutations.
These genes are often difficult to target (chromatin modifiers, ubiquitin ligases); but that's changing with the increasing ease of binder design with all kinds of properties that make the "undruggable" druggable.
Exciting times to be alive in Psychiatry.
Source: Sealock et al. bioRxiv 2026 — doi.org/10.64898/2026.…
1
3
20
1,526