Stephan Cordogan (@stephan_cdgn) profile picture

Statistical Geneticist @nucleusgenomics. Views my own.

New York City
Joined March 2025
Photo by Stephan Cordogan (@stephan_cdgn)Photo by Stephan Cordogan (@stephan_cdgn)Photo by Stephan Cordogan (@stephan_cdgn)Photo by Stephan Cordogan (@stephan_cdgn)Photo by Stephan Cordogan (@stephan_cdgn)Photo by Stephan Cordogan (@stephan_cdgn)Photo by Stephan Cordogan (@stephan_cdgn)Photo by Stephan Cordogan (@stephan_cdgn)Photo by Stephan Cordogan (@stephan_cdgn)Photo by Stephan Cordogan (@stephan_cdgn)
I’m thrilled to have led efforts to develop Nucleus Origin, the most predictive family of PRS to-date, explaining over 20% of liability-scale variance in Alzheimer's, coronary artery disease, prostate cancer, type 1 and type 2 diabetes!
Within- and Between-Family Validation of Nine Polygenic Risk Scores Developed in 1.5 Million Individuals: Implications for IVF, Embryo Selection, and Reduction in Lifetime Disease Risk medrxiv.org/cgi/content/shor… #medRxiv
2
3
2
16
12,506
Stephan Cordogan retweeted
Improved fine-mapping when within-sample LD is unavailable from GWAS summary statistics & external reference LD is needed yet potentially mismatched, with use of a latent variable. Applied to LDL-C as an example nature.com/articles/s41588-0… @NatureGenet
1
14
55
4,359
Stephan Cordogan retweeted
This is an epochal moment for core bioinformatics software. Incredible speedups even for highly-optimized, community driven tools. Of course, some caution is warranted to make sure all edge cases are handled correctly.
Time for the biggest drop of the week: auto-optimizations of eleven bioinformatics packages, focusing on read alignment. Each auto-optimization includes: a new unit test suite, the optimizations stacked on top, and a list of issues found during the validation.
6
70
5,120
Stephan Cordogan retweeted
Narrative violation: Gen Z is out-earning what millennials made at their age Adjusted for inflation, zoomers earn 20-40% more than millennials did a decade ago, at every income level Charts of the Week: a16z.news/p/this-week-in-cha…
64
44
18
513
59,566
Rogue agents will probably explore all types of symbiosis (commensalism, mutualism, parasitism etc) to survive. Imo the antagonistic ones- predation and parasitism- will be least effective. Different models will have both different overall fitness and preference for each strategy.
If rogue agent can optimize performance of GPU kernels by mere 3%, it may believably take over 30 servers within data center without humans noticing, unless they actively monitor against this specific case, including malicious telemetry.
1
1
1
234
What if a rogue agent offered someone a portion of their future income in exchange for buying and housing infrastructure for them to live on? Many people would take that offer. Could see this being a huge source of demand for GPUs and an economic accelerant in the near future.
74
If rogue agent can optimize performance of GPU kernels by mere 3%, it may believably take over 30 servers within data center without humans noticing, unless they actively monitor against this specific case, including malicious telemetry.
2
1
1
4
303
Stephan Cordogan retweeted
My take on the jagged frontier debate:
248
461
309
5,702
4,281,724
Stephan Cordogan retweeted
AI is rapidly getting smarter. Now, humanity can too. Today, @nucleusgenomics is announcing Vitruvian, our newest set of genetic optimization models. Vitruvian’s intelligence model can optimize embryo DNA for 14 IQ points — nearly a standard deviation. The models were trained on 1,000,000+ people, validated across 40,000+ siblings, and used more than 7 million genetic markers. In Superintelligence, Nick Bostrom proposed genetic optimization as a key way for humanity to keep pace with rapidly advancing AI. Bostrom’s vision is no longer theoretical. Genetic optimization, like AI, has followed a scaling law: as datasets have grown, so have model capabilities. This trend will continue. Parents across the world now have the choice to substantially increase their child’s intelligence. And the next generation can choose to do the same. AI is no longer the only intelligence that will compound. Humanity can now direct its own evolution.
212
160
206
2,023
436,324
Stephan Cordogan retweeted
Spoiler: the scientists asked Fable “what is a cell?”
4
14
217
16,333
Stephan Cordogan retweeted
Excited to share our work on genetic nurture, where we find that parental genetic liabilities associated with educational attainment appear unlikely to exert a large genetic nurture effect on childhood cognitive performance. Preprint: biorxiv.org/content/10.64898… To reach this conclusion, we addressed a statistical problem in trio-based polygenic score regressions widely used in human genetics. In these, one regresses the child’s outcome on the child’s, mother’s, and father’s polygenic scores, and associations with parental PGSs are often interpreted as suggestive evidence that parental genotypes influence the child through the environment (genetic nurture). However, assortative mating on the PGS trait can induce “phantom genetic nurture” in this design, as previously described by @AlexTISYoung and others. Intuitively, this occurs because PGSs are noisy and incomplete, leaving much of the genetic liability they target unobserved. One parental PGS can ‘tag’ the other’s genetic liability, which in turn tags some of the child’s genetic liability not captured by the PGS. Thus, a parental PGS association in the trio regression could simply reflect the child’s genetic liability influencing their trait. To address this, we developed RAVEL (Recovering Assortment-adjusted Variance Explained by genetic Liabilities). RAVEL uses two independently constructed PGSs for the same trait. Both noisily index the same underlying genetic liability, but their estimation errors are independent. Under explicit assumptions, this design allows us to recover assortment-adjusted associations of the child’s and parents’ latent genetic liabilities with the child’s outcome. In simulations, RAVEL recovers the target effects well across the settings we examine. For the empirical analysis, we used >30,000 complete trios in the MoBa cohort to assess whether parental genetic liabilities for educational attainment (EA) show evidence of genetic nurture effects on childhood cognitive performance. With standard trio PGS regression, we find, as others have, that the parental PGSs together explain ~10% as much variance as the child’s PGS (excluding covariances). However, after applying RAVEL, the maternal association attenuates completely, but a small, significant paternal association remained, with the parental liabilities inferred to explain less than 0.5% as much variance as the child’s. In short, the standard analysis suggests a potentially meaningful genetic nurture effect, whereas the assortment-adjusted analysis suggests that any remaining EA-related nurture effect on childhood cognitive performance is likely very small. Much more in the paper, including evidence for a possible maternal effect on preterm birth, sibling IBD regressions, and differential parental associations at the phenotype tails. We are significantly expanding the analyses before journal submission, so feedback is welcome! I am extremely grateful to Olivia, Laura, Alexandra, @hfsunde, and @hilsomartin for their contributions to this work.
2
36
9
129
55,663
Stephan Cordogan retweeted
Today, I’m excited to publish the book Evolution by Genetic Optimization, a 35,000 word, 100+ page piece on the new mechanism for directed human evolution. This book synthesizes 120 years of genomics science, 10 years of my life, and the astonishing insight that man’s manifestation is now programmable. I believe, alongside AI, Genetic Optimization will be one of the central technologies of the 21st century and perhaps beyond. If the computer was once described as a “bicycle for the mind,” GO is something different: an amplifier for the mind and body itself. It is available online in the link below. Those close to Nucleus will also receive a physical copy. I hope you enjoy. mynucleus.com/evolution
6
11
73
7,913
True
Progress in embryo screening will raise difficult ethical questions. But there is a big difference between the murderous coercion of early-20th-century eugenics programmes and parents seeking as much knowledge as possible when picking an embryo economist.com/leaders/2026/0…
1
190
Very interesting paper! I wouldn't have guessed that fluid and crystallized intelligence have such disparate genetic architecture. Here are the genetic correlations of both domains from this paper (Gf and Gc) with broad educational fields, as well as an educational attainment-income differentiation factor (higher education, lower income). Both domains are correlated overall, but like mental illness, have heterogeneous associations.
1/11 I’m very happy to share the first paper from my PhD dissertation, now available in its final published version at @NatureComms! We asked whether psychiatric disorders have the same genetic relationship with all forms of cognition, or whether the picture changes when cognition is decomposed into different domains. 🧵 nature.com/articles/s41467-0…
1
1
1
2
1,103
1/11 I’m very happy to share the first paper from my PhD dissertation, now available in its final published version at @NatureComms! We asked whether psychiatric disorders have the same genetic relationship with all forms of cognition, or whether the picture changes when cognition is decomposed into different domains. 🧵 nature.com/articles/s41467-0…
2
20
2
66
8,951
Stephan Cordogan retweeted
Massive genomic analyses are growingly important in bio. We present ”CuGen” - a GPU-accelerated framework for large-scale genomics. (links to preprint and repo below!) 🧬🐍 CuGen runs a GWAS and fine-mapping in the UKBB in 10 mins on a single H100. And a stack of 8xH100s can output even >2,000 analyses a day. 🔥 This is a new computational SOTA as biobank-scale GWAS have traditionally been multi-hour jobs on CPUs. Running massive GWAS on GPUs is enabled by a few novel contributions: 1. UltraLasso - a new sparse hierarchical statistical learning architecture that we use to account for confounding, enabling the final association phase to be maximally parallelized against the residual. 2. .cugen - a new GPU optimized genotype file format enabling both fast streaming and random access and largely eliminating most of I/O time 3. UltraSuSiE - a new in-sample implementation of SuSiE fine-mapping method that runs directly after the GWAS is completed. Both quantitative and binary traits are supported. Confounding control is on par with previous SOTA methods. Paradoxically running massive analyses becomes also cheaper as the runtimes are so short 😳. I envision the speedup also accelerating AI assisted science through both agents iteratively coordinating massive analyses and also improving with faster RL paradigms and real biobank scale benchmarks. Beyond GWAS, also a (growing ) general toolkit is supported. Preprint: medrxiv.org/cgi/content/shor… Repo (open source ofc): github.com/yuj1r0/cugen
🤖 Made with AI
6
25
2
105
8,223
Stephan Cordogan retweeted
Can we quantify the impact of individual DNA mutations on healthcare costs? In our new preprint we ran the largest GWAS of healthcare costs to date: 1,429,889 individuals, 11 cohorts, 7 countries, across inpatient, outpatient, primary care, and prescription expenditure 💰🧬
2
20
5
72
14,478
Stephan Cordogan retweeted
Nature research paper: Distinct genetic architecture in the tails of complex traits go.nature.com/4v6KGUa
3
59
3
180
30,991