Shai Carmi
@shaicarmi.bsky.social
Associate professor at the Hebrew University of Jerusalem.
Statistical, population, and medical genetics; preimplantation genetic testing. Views my own.
http://scarmilab.org
Statistical, population, and medical genetics; preimplantation genetic testing. Views my own.
http://scarmilab.org
More details emerge on this affair.
1. A paper has been submitted to a "high profile" journal. I think no editor in their right mind should take it, but I have no hope.
2. Details on the psychiatric PRS analysis were reported by the investigator.
edition.cnn.com/2025/11/13/s...
1. A paper has been submitted to a "high profile" journal. I think no editor in their right mind should take it, but I have no hope.
2. Details on the psychiatric PRS analysis were reported by the investigator.
edition.cnn.com/2025/11/13/s...
November 14, 2025 at 9:13 AM
More details emerge on this affair.
1. A paper has been submitted to a "high profile" journal. I think no editor in their right mind should take it, but I have no hope.
2. Details on the psychiatric PRS analysis were reported by the investigator.
edition.cnn.com/2025/11/13/s...
1. A paper has been submitted to a "high profile" journal. I think no editor in their right mind should take it, but I have no hope.
2. Details on the psychiatric PRS analysis were reported by the investigator.
edition.cnn.com/2025/11/13/s...
The results show, consistently with previous studies, a significant but very small improvement in AUC (Delta~0.01).
November 7, 2025 at 8:33 AM
The results show, consistently with previous studies, a significant but very small improvement in AUC (Delta~0.01).
Interesting to also see this paper, attempting to predict 10-year myocardial infarction using both PRS and non-genetic (NG) data.
www.nature.com/articles/s43...
The NG data was either established risk factors or learned from a high-dim dataset. Prediction used logistic regression/neural network.
www.nature.com/articles/s43...
The NG data was either established risk factors or learned from a high-dim dataset. Prediction used logistic regression/neural network.
November 7, 2025 at 8:33 AM
Interesting to also see this paper, attempting to predict 10-year myocardial infarction using both PRS and non-genetic (NG) data.
www.nature.com/articles/s43...
The NG data was either established risk factors or learned from a high-dim dataset. Prediction used logistic regression/neural network.
www.nature.com/articles/s43...
The NG data was either established risk factors or learned from a high-dim dataset. Prediction used logistic regression/neural network.
The actual numbers show that all health outcomes improved, with some outcomes very close to significance. So the headline is a bit misleading. But the effect sizes are nevertheless very small, particularly that they are likely to diminish over time.
November 7, 2025 at 8:33 AM
The actual numbers show that all health outcomes improved, with some outcomes very close to significance. So the headline is a bit misleading. But the effect sizes are nevertheless very small, particularly that they are likely to diminish over time.
At the bone marrow, triploidy provided the child with a "natural gene therapy", because the triploid cells had two X chrs, including one not carrying the mutation.
This gave the triploid lineage a growth advantage, and over time, it took over the bone marrow and rescued hematopoiesis.
This gave the triploid lineage a growth advantage, and over time, it took over the bone marrow and rescued hematopoiesis.
November 5, 2025 at 8:04 AM
At the bone marrow, triploidy provided the child with a "natural gene therapy", because the triploid cells had two X chrs, including one not carrying the mutation.
This gave the triploid lineage a growth advantage, and over time, it took over the bone marrow and rescued hematopoiesis.
This gave the triploid lineage a growth advantage, and over time, it took over the bone marrow and rescued hematopoiesis.
Turns out, as an early embryo, one of his cells incorporated the second polar body. That cell became triploid, carrying two copies of the maternal genome.
The embryo retained this lineage and continued to develop with low frequency mosaic triploidy.
The embryo retained this lineage and continued to develop with low frequency mosaic triploidy.
November 5, 2025 at 8:04 AM
Turns out, as an early embryo, one of his cells incorporated the second polar body. That cell became triploid, carrying two copies of the maternal genome.
The embryo retained this lineage and continued to develop with low frequency mosaic triploidy.
The embryo retained this lineage and continued to develop with low frequency mosaic triploidy.
Celebrating the release of the hostages in my neighborhood and all over Israel.
After two years, we can breathe again.
Let's hope this day follows with peace and prosperity to the whole region.
After two years, we can breathe again.
Let's hope this day follows with peace and prosperity to the whole region.
October 13, 2025 at 11:12 AM
Celebrating the release of the hostages in my neighborhood and all over Israel.
After two years, we can breathe again.
Let's hope this day follows with peace and prosperity to the whole region.
After two years, we can breathe again.
Let's hope this day follows with peace and prosperity to the whole region.
Using 30k sibling pairs from Mexico city, the authors find that indigenous American ancestry is associated with lower height (-1.5SD) and higher type 2 diabetes risk
There were no associations with other traits, including education.
3/3
There were no associations with other traits, including education.
3/3
September 11, 2025 at 5:57 AM
Using 30k sibling pairs from Mexico city, the authors find that indigenous American ancestry is associated with lower height (-1.5SD) and higher type 2 diabetes risk
There were no associations with other traits, including education.
3/3
There were no associations with other traits, including education.
3/3
Here, the authors suggest to study pairs of siblings in admixed populations.
Due to the randomness of Mendelian inheritance, sibs differ in their ancestry proportions. At the same time, they have a nearly identical environment.
2/3
Due to the randomness of Mendelian inheritance, sibs differ in their ancestry proportions. At the same time, they have a nearly identical environment.
2/3
September 11, 2025 at 5:57 AM
Here, the authors suggest to study pairs of siblings in admixed populations.
Due to the randomness of Mendelian inheritance, sibs differ in their ancestry proportions. At the same time, they have a nearly identical environment.
2/3
Due to the randomness of Mendelian inheritance, sibs differ in their ancestry proportions. At the same time, they have a nearly identical environment.
2/3
We showed (mathematically and numerically) that using a fixed number of births (e.g., the mean number of births per IVF cycle) still overestimates the risk reduction (see Figure). It is important to account for the randomness of embryo transfer outcomes.
September 10, 2025 at 3:28 PM
We showed (mathematically and numerically) that using a fixed number of births (e.g., the mean number of births per IVF cycle) still overestimates the risk reduction (see Figure). It is important to account for the randomness of embryo transfer outcomes.
Parameters the user can control include:
* Disease prevalence
* PRS accuracy
* Number of embryos
* Live birth rate
* Parental PRS
* Disease status of parents and previously born siblings
* Embryo selection strategy
* Disease prevalence
* PRS accuracy
* Number of embryos
* Live birth rate
* Parental PRS
* Disease status of parents and previously born siblings
* Embryo selection strategy
September 10, 2025 at 3:28 PM
Parameters the user can control include:
* Disease prevalence
* PRS accuracy
* Number of embryos
* Live birth rate
* Parental PRS
* Disease status of parents and previously born siblings
* Embryo selection strategy
* Disease prevalence
* PRS accuracy
* Number of embryos
* Live birth rate
* Parental PRS
* Disease status of parents and previously born siblings
* Embryo selection strategy
To address that, we developed an online app, called PEStimate.
Users specify the parameters of the embryos, the disease, and the family, and the app computes the risk of the child with or without PES.
You can also generate plots of the risk reductions vs key parameters.
Users specify the parameters of the embryos, the disease, and the family, and the app computes the risk of the child with or without PES.
You can also generate plots of the risk reductions vs key parameters.
September 10, 2025 at 3:28 PM
To address that, we developed an online app, called PEStimate.
Users specify the parameters of the embryos, the disease, and the family, and the app computes the risk of the child with or without PES.
You can also generate plots of the risk reductions vs key parameters.
Users specify the parameters of the embryos, the disease, and the family, and the app computes the risk of the child with or without PES.
You can also generate plots of the risk reductions vs key parameters.
August 13, 2025 at 10:47 AM
More than 2000 papers are now published each year on polygenic scores.
Is that because polygenic scores are so important for precision medicine?
Or because polygenic score studies became so easy to conduct?
Is that because polygenic scores are so important for precision medicine?
Or because polygenic score studies became so easy to conduct?
August 1, 2025 at 12:29 PM
More than 2000 papers are now published each year on polygenic scores.
Is that because polygenic scores are so important for precision medicine?
Or because polygenic score studies became so easy to conduct?
Is that because polygenic scores are so important for precision medicine?
Or because polygenic score studies became so easy to conduct?
However, I think confounders can be tricky to avoid, particularly balancing of offspring sex. For example, when considering only the first two children of each women, the distribution was binomial.
July 21, 2025 at 9:10 PM
However, I think confounders can be tricky to avoid, particularly balancing of offspring sex. For example, when considering only the first two children of each women, the distribution was binomial.
An interesting article suggesting that *maternal* (genetic and other) factors affect the sex of the offspring of a woman. The distribution of the number of male/female children is beta-binominal, implying the existence of woman-specific offspring sex probability.
www.science.org/doi/10.1126/...
www.science.org/doi/10.1126/...
July 21, 2025 at 9:10 PM
An interesting article suggesting that *maternal* (genetic and other) factors affect the sex of the offspring of a woman. The distribution of the number of male/female children is beta-binominal, implying the existence of woman-specific offspring sex probability.
www.science.org/doi/10.1126/...
www.science.org/doi/10.1126/...
Two useful new reviews on ancestry inference in population genetics:
onlinelibrary.wiley.com/doi/10.1111/...
genomebiology.biomedcentral.com/articles/10....
onlinelibrary.wiley.com/doi/10.1111/...
genomebiology.biomedcentral.com/articles/10....
July 17, 2025 at 2:57 PM
Two useful new reviews on ancestry inference in population genetics:
onlinelibrary.wiley.com/doi/10.1111/...
genomebiology.biomedcentral.com/articles/10....
onlinelibrary.wiley.com/doi/10.1111/...
genomebiology.biomedcentral.com/articles/10....
An impressive study of UCLA's ATLAS biobank.
94k genotyped, 62k exomes. Paper includes fine-scale IBD-based clustering, disease burden and rare variants per cluster, GWAS on 776 phenotypes, and rare variant testing, discovering PTPRU as linked to GLP-1 response.
www.medrxiv.org/content/10.1...
94k genotyped, 62k exomes. Paper includes fine-scale IBD-based clustering, disease burden and rare variants per cluster, GWAS on 776 phenotypes, and rare variant testing, discovering PTPRU as linked to GLP-1 response.
www.medrxiv.org/content/10.1...
June 15, 2025 at 9:44 AM
An impressive study of UCLA's ATLAS biobank.
94k genotyped, 62k exomes. Paper includes fine-scale IBD-based clustering, disease burden and rare variants per cluster, GWAS on 776 phenotypes, and rare variant testing, discovering PTPRU as linked to GLP-1 response.
www.medrxiv.org/content/10.1...
94k genotyped, 62k exomes. Paper includes fine-scale IBD-based clustering, disease burden and rare variants per cluster, GWAS on 776 phenotypes, and rare variant testing, discovering PTPRU as linked to GLP-1 response.
www.medrxiv.org/content/10.1...
Mefford et al considers quantiles of (non-PRS) risk, which can point to heterogeneity in the PRS predictive value without requiring the specification of the background variables.
Figures: height and BMI show little or substantial heterogeneity, respectively.
www.sciencedirect.com/science/arti...
Figures: height and BMI show little or substantial heterogeneity, respectively.
www.sciencedirect.com/science/arti...
June 8, 2025 at 9:21 AM
Mefford et al considers quantiles of (non-PRS) risk, which can point to heterogeneity in the PRS predictive value without requiring the specification of the background variables.
Figures: height and BMI show little or substantial heterogeneity, respectively.
www.sciencedirect.com/science/arti...
Figures: height and BMI show little or substantial heterogeneity, respectively.
www.sciencedirect.com/science/arti...
Interesting to see two new papers reporting variable accuracy of polygenic scores across non-genetic (or at least non-PRS) backgrounds.
Cudic et al considers intersections of age, sex, smoking, alcohol, income, and deprivation.
www.medrxiv.org/content/10.1...
Cudic et al considers intersections of age, sex, smoking, alcohol, income, and deprivation.
www.medrxiv.org/content/10.1...
June 8, 2025 at 9:21 AM
Interesting to see two new papers reporting variable accuracy of polygenic scores across non-genetic (or at least non-PRS) backgrounds.
Cudic et al considers intersections of age, sex, smoking, alcohol, income, and deprivation.
www.medrxiv.org/content/10.1...
Cudic et al considers intersections of age, sex, smoking, alcohol, income, and deprivation.
www.medrxiv.org/content/10.1...
Nice method using graph theory to take an IBD network and convert it into per-individual continuous ancestry components. These "spectral" components outperform PCA for correction of population structure in GWAS.
www.medrxiv.org/content/10.1...
www.medrxiv.org/content/10.1...
June 8, 2025 at 8:36 AM
Nice method using graph theory to take an IBD network and convert it into per-individual continuous ancestry components. These "spectral" components outperform PCA for correction of population structure in GWAS.
www.medrxiv.org/content/10.1...
www.medrxiv.org/content/10.1...
Interesting commentary in Cell. In 9,251 publicly available human metagenomes, they found reads mapping to human mtDNA in ~20% of the datasets. In almost all it was possible to derive the haplogroup.
This is an important privacy gap that should be closed.
www.cell.com/cell/fulltex...
This is an important privacy gap that should be closed.
www.cell.com/cell/fulltex...
May 17, 2025 at 9:19 AM
Interesting commentary in Cell. In 9,251 publicly available human metagenomes, they found reads mapping to human mtDNA in ~20% of the datasets. In almost all it was possible to derive the haplogroup.
This is an important privacy gap that should be closed.
www.cell.com/cell/fulltex...
This is an important privacy gap that should be closed.
www.cell.com/cell/fulltex...
OK, how does that work? Of course, it perfectly distinguishes cases and controls in the family! (red and green triangles; the blue dots are not described - probably population samples)
I haven't seen any evidence for out-of-sample validation, so this is probably overfitting.
I haven't seen any evidence for out-of-sample validation, so this is probably overfitting.
May 17, 2025 at 9:09 AM
OK, how does that work? Of course, it perfectly distinguishes cases and controls in the family! (red and green triangles; the blue dots are not described - probably population samples)
I haven't seen any evidence for out-of-sample validation, so this is probably overfitting.
I haven't seen any evidence for out-of-sample validation, so this is probably overfitting.
They doubled the weight of risk alleles that were more common in the family's cases than controls, and halved the weight of risk alleles that were more common in the controls.
They also added 1 for one of the sexes (no details).
They also added 1 for one of the sexes (no details).
May 17, 2025 at 9:09 AM
They doubled the weight of risk alleles that were more common in the family's cases than controls, and halved the weight of risk alleles that were more common in the controls.
They also added 1 for one of the sexes (no details).
They also added 1 for one of the sexes (no details).
So far so good. But...
The PRS was developed based on an existing GWAS and validated in the UK biobank. However, that PRS did not predict disease status in the father's family (only reported in the Discussion! see excerpt).
So the researchers modified the PRS...
The PRS was developed based on an existing GWAS and validated in the UK biobank. However, that PRS did not predict disease status in the father's family (only reported in the Discussion! see excerpt).
So the researchers modified the PRS...
May 17, 2025 at 9:09 AM
So far so good. But...
The PRS was developed based on an existing GWAS and validated in the UK biobank. However, that PRS did not predict disease status in the father's family (only reported in the Discussion! see excerpt).
So the researchers modified the PRS...
The PRS was developed based on an existing GWAS and validated in the UK biobank. However, that PRS did not predict disease status in the father's family (only reported in the Discussion! see excerpt).
So the researchers modified the PRS...