Tony

We’re excited to share our latest publication in @CellGenomics: “Streamlining Large-Scale Genomic Data Management: Insights from the UK Biobank Whole-Genome Sequencing Data”. Sincerely thanks to @drarwood, @muzizimumu1, @XihongLin, Yuxin Yuan, Gareth Hawkes, Robin Beaumont, Michael Weedon, all collaborators and study participants from the @uk_biobank Program. We highlight the annotated Genomic Data Structure (aGDS) format, the vcf2agds toolkit, and the STAARpipeline, which together: • Seamlessly integrate genotypes + functional annotations in an all-in-one compact file for downstream analyses 🗃️ • Reduce UKB 500k WGS storage from 1473.85 TiB (GraphTyper pVCF, #23374) to 1.10 TiB (aGDS); and from 17.87 TiB (ML-Corrected DRAGEN pVCF, #24311) to 1.65 TiB (aGDS) 🗄️ • Enable scalable, functionally informed WGS association analyses across coding & noncoding genome 🧬 • Empower open-source, RAP-integrated genomic analyses for hundreds of thousands of samples ☁️ Applying STAARpipeline to the UKB 500k WGS data for total cholesterol, we identified 480 genome-wide significant rare variant associations, including 200 coding and 280 noncoding functional categories. These signals encompass lipid biology mainstays (e.g., PCSK9, APOB, NPC1L1, LDLR, APOE) and regulatory variants mapped to promoter, enhancer, and UTR regions, demonstrating the power of biobank-scale WGS for genomic discovery. All tools are open source and freely available: 🔗 vcf2agds toolkit: github.com/drarwood/vcf2a… 🔗 STAARpipeline: github.com/xihaoli/STAARp… 🔗 See the paper for complete links to software repos: cell.com/cell-genomics/…

So excited to share my first PhD paper, discussing GWAS methodologies in multi-ancestry contexts: To pool or not to pool? Very proud of this work and grateful for the supervision of @AndrewHaoyu and @GENES_PK! Read more: cell.com/ajhg/abstract/…

Excited to share our new manuscript on evaluating multi-ancestry GWAS methods using large-scale simulations & real data from UK Biobank (N≈324K) & All of Us (N≈207K)! My first thesis paper under the guidance of @GENES_PK & @AndrewHaoyu. Available at: medrxiv.org/content/10.110…

Please join us tomorrow at HSPH for a @HarvardPqg Working Group seminar with @kunhsingyu , who will be discussing AI foundation models in pathology!

📢Out now! @xihaoli, @pnatarajanmd, @muzizimumu1, @ZHLiu13, @XihongLin, and colleagues from @nih_nhlbi present a framework for functionally-informed multi-trait rare variant analysis of biobank-scale sequencing studies. nature.com/articles/s4358… 🔓rdcu.be/d82Fx

Please join us for our last @HarvardPqg Working Group seminar of 2024, next Tuesday with @yk_tani presenting his exciting recent work on polygenic prediction!

Just in time! Come join us with @saorisakaue on 11/19 for her recent work revealing genetic regulatory mechanisms through eQTL causal variants! @HarvardBiostats @HarvardEpi

Our paper introducing a PRS-based translation roadmap for lung cancer care is now published on @eBioMedicine ! Wonderful collaboration with @AndrewHaoyu , @lishiunchen , ILCCO and GSCAN towards the clinical implementation of PRS. sciencedirect.com/science/articl…

🚀 Excited to share our new preprint on RICE, a PRS framework that integrates common and rare variants to enhance genetic risk prediction across diverse ancestries! Led by my postdoc fellow Jacob Williams, co-mentored with @xihaoli. Check it out here: medrxiv.org/content/10.110… 1/N

Check out the presentations from our lab and our close collaborators #ASHG2024. Such a hard-working year. So proud of the group!!! Very humbled to be able to work with so many great collaborators @xihaoli @XihongLin @GENES_PK @Diptavo @HongyuZhao2 @Tony_Chen6 @AlexiaDiasF

Just in time, I'll be presenting this work today at the @HarvardPqg conference in Boston, and at the @GeneticsSociety conference in Denver. Looking forward to comments, feedback, and discussion!

SPLENDID enables computation of a single, label-free model with robust prediction across all ancestries, which we hope can enhance fairness and interpretability in clinical implementation. Software and tutorials are available on GitHub. github.com/chen-tony/SPLE…

Using training data from All of Us and validation in UK Biobank, SPLENDID outperforms existing methods, particularly in non-European ancestry, selects fewer variants, remains robust to ancestry proportions in tuning, and can be used with only training data.

SPLENDID uses an ensemble group-L0L2 penalized regression framework to model interactions between genetic variants and ancestry PCs. This allows for simultaneous estimation of shared and heterogeneous effects across diverse populations without discrete ancestry labels.

Excited to share our preprint for SPLENDID, a new PRS method using continuous ancestry interactions and biobank-scale individual level data. Many thanks to @AndrewHaoyu , Rahul Mazumder, and @XihongLin for their valuable guidance on this work! biorxiv.org/content/10.110…

Excited to have you @Aoxing2 , really looking forward to your talk!!

Our new paper on semi-supervised machine learning method for predicting homogeneous ancestry groups to assess HWE in diverse whole #genome sequencing studies. The work was motivated by NHGRI Genome Sequencing Program - CCDG data. Applied to HWE QC of 60,000+ diverse whole genomes

@AndrewHaoyu @XihongLin Thank you so much Haoyu!! Extremely grateful to have had you as a mentor from the beginning 🙏

@Tony_Chen6 is a very talented student with a great passion for science! I always feel privileged to work with Tony. Super grateful for the opportunity to co-mentor PhD students with @XihongLin and have learned a lot from her mentoring style over the years.🫡

Congratulations to @Tony_Chen6 on publishing his first thesis paper🥳🥳🥳! A new PRS method named All-Sum, using L0L2 penalty, shows impressive prediction, a very sparse PRS SNP set, and excellent computation speed. Great collaboration with Rahul and @XihongLin! @NCIEpiTraining

ALL-Sum combines fast optimization of L0L2 penalized regression with ensemble learning to improve PRS accuracy and computational efficiency vs. many popular methods. Software and accompany data are all available on GitHub (github.com/chen-tony/ALL-…)

Our paper introducing a fast and scalable PRS method (ALL-Sum) is now available on @PNASNews! Many thanks to @AndrewHaoyu, Rahul Mazumder, and @XihongLin for their mentorship, as well as @BhramarBioStat and @HongyuZhao2 for reviewing! pnas.org/doi/10.1073/pn…