Liaoyi Xu
@xliaoyi
PhD student @UTAustin; previous @ZJU_China
gsMap paper online. Source code: github.com/JianYang-Lab/g…. Results visualisation website: yanglab.westlake.edu.cn/gsmap.
Nature research paper: Spatially resolved mapping of cells associated with human complex traits go.nature.com/4c3EAvv
My first textbook is officially available for pre-order today. Grab a copy if you're interested in developing a (statistical) research-oriented mindset. The book is designed as a stepping stone from statistical coursework to novel statistical research. routledge.com/Exercises-in-S…
I am thrilled to share one of the most important projects from my PhD! We developed the method JOBS (JOint model of Bk-eQTLs as a weighted sum of Sc-eQTLs) and applied it to blood bulk (eQTLGen) and sc-eQTL (OneK1K). 1/3
I am excited to share our new work that generates a single cell eQTL atlas for immune cell types. The work integrates bulk eQTLs with sc-eQTLs, which boosts the power for identifying sc-eQTLs for up to 4 folds. As such, we have the power equivalent to sc-RNASeq of 4K individuals.
Hong Wang delivered her lecture “Kakeya Sets in R^3” to a packed crowd at yesterday’s Mathematics Colloquium!
Last year, I joined DeepSeek with no RL experience. While conducting Mathshepherd and DeepSeekMath research, I independently derived this unified formula to understand various training methods. It felt like an "aha moment", though I later realized it was PG.
If you can only read one DeepSeek paper in your life, read DeepSeek Math. Everything else is either ≈obvious in hindsight or clever optimization. DeepSeek Math is a tour de force of data engineering, general DL LLM methodology, RL, and just beautiful. Just 22 pages.
In my latest paper (with M. Hooten and @vagheesh), we developed a dyadic model that accounts for the flow of biological processes in space-time. We applied our method to ancient human DNA data to study human movement in Bronze Age Europe. (1/5) academic.oup.com/biometrics/art…
When in evolutionary time does accelerated genomic change occur across phenotypes? Newly published today in @CellGenomics , my latest manuscript with @mashaals and @vagheesh integrates comparative, functional, and ancient genomic data with GWAS. Read more: cell.com/cell-genomics/…
Scientists from @UTAustin & @UCLA used ancient DNA samples to look back in time & reveal new insights into how ancient Europeans adapted to their environments over 7,000 years of European history. #AncientDNA @UT_Stats @Texas_IB @Vagheesh @DevPandeya cns.utexas.edu/news/research/…
New paper led by star PhD student @mrschwob (who is currently on the statistics job market) building a bayesian hierarchical model to understand gene flow across space and time in ancient DNA datasets. academic.oup.com/biometrics/art…
Thrilled to share that the first project from my PhD is published in Nature Communications (@NatureComms) ! 🎉 We explored Holocene European aDNA to uncover how selective sweeps have been obscured by neutral processes like drift and admixture. Read more: nature.com/articles/s4146…
In this preprint, Aylwyn Scally, Richard Durbin and I present evidence for deep population structure shared by all modern humans biorxiv.org/content/10.110… @aylwyn_scally @richard_durbin
Our new AI paper reveals surprising geometric structure in the LLM-learned concepts: 1) They form brain-like "lobes", 2) they form "semantic crystals" much more precise than it first seems, and 3) the concept cloud is more fractal than round:
1/6 New paper! “The Geometry of Concepts: Sparse Autoencoder Feature Structure.” We find that the concept universe of SAE features has interesting structure at three levels: 1) “atomic” small-scale, 2) “brain” intermediate-scale, and 3) “galaxy” large-scale!
Excited to share my 1st-author work from the @arbelharpak Lab! We examine the accuracy of polygenic score (PGS) predictions across individuals in the UK Biobank. We make 3 observations that expose gaps in our understanding of PGS “portability”. biorxiv.org/content/10.110… (1/28)
Overall saw some great talks, met awesome folks in the #SMBE community, and enjoyed the food and culture of Puerto Vallarta. Thanks to all the SMBE organizers for their work and looking forward to the next one!
@devpandeyay and @xliaoyi also presented posters: “Leveraging ancient DNA to uncover signals of natural selection in Europe lost to admixture or drift” and “The genetic architecture and consequences of the human pelvic form”. Check out what the @vagheesh Narasimhan lab is up to!
Had a great first time at #SMBE2024! Was thrilled and grateful to present my work on “The trait specific timing of accelerated genomic change in the human lineage” which also has a preprint available at doi.org/10.1101/2022.0… @mashaals @vagheesh
A new release of the HARE pipeline for enrichment analysis (with lots of new functionality!) is live and published in @JOSS_TheOJ joss.theoj.org/papers/10.2110…. Cannot thank @EucharistKun and @vagheesh enough for all their work making v1.2 happen.
The main issue testing these classic theories in human evolution is sample size. Using AI to measure skeletal phenotypes in >30,000 people, we increased sample size by ~2 orders of magnitude. I think our results shed new light on this 6 decade old question x.com/xliaoyi/status…
The obstetrical dilemma - a 60yr old hypothesis in evolutionary theory, suggests that a narrow pelvis aids bipedal locomotion but complicates childbirth. But how does it stand up to modern empirical evidence? Read our preprint to find out! biorxiv.org/content/10.110…