#statgen2024 — Public Fediverse posts

STATGEN 2024 talk
A Kernel-Based Neural Network for High-dimensional Risk Prediction on Massive Genetic Data
Qing Lu

Neural Network
Nonlinear
Non-additive

Kernel-Based Neural Network (KNN)
kernel matrics constructed based on the genetic variables.

Related preprint:
An Association Test Based on Kernel-Based Neural Networks for Complex Genetic Association Analysis
https://arxiv.org/abs/2312.06669

1/

#STATGEN2024 #Genetics #StatisticalGenetics

STATGEN 2024 talk
Improved methods for empirical Bayes multivariate multiple testing and effect size estimation
Yunqi Yang

Empirical Bayes multivariate normal means (EBMNM) model [Urbut et al., 2019]

Allow for heterogeneous sharing of eQTLs in multiple tissues (e.g., some are shared across all tissues, some are shared only within brain tissues, etc.)

Truncated Eigenvalue Decomposition

udr: Ultimate Deconvolution in R
https://stephenslab.github.io/udr/

#STATGEN2024 #Genetics #StatisticalGenetics

STATGEN 2024 talk
MultiSTAAR: A statistical framework for powerful multi-trait rare variant analysis in large-scale whole-genome sequencing studies
Xihao Li

Functionally-informed Multi-Trait MultiSTAAR approach.

MultiSTAAR-O: Omnibus test
1. Burden
2. SKAT
3. ACAT-V

Li X et al. A statistical framework for powerful multi-trait rare variant analysis in large-scale whole-genome sequencing studies. bioRxiv doi: 10.1101/2023.10.30.564764.

#STATGEN2024 #Genetics #StatisticalGenetics

STATGEN 2024 talk
Adventures in Human Genetics: Purpose, Serendipity, Innovation
Gonçalo Abecasis

"It is important to think carefully about what is the right question, and what are the right statistics. But there is a lot of opportunity in thinking about what is the best design to answer the question."

Goal
Understand disease
Treat
Predict disease
Prevent

Can learn from natural experiments in millions of people.

1/

#STATGEN2024 #Genetics #StatisticalGenetics

STATGEN 2024 talk
Working towards Inclusivity in Genetic Studies: Estimating accurate population structure with Small Reference Sample Sizes
Souha Tifour

Arriaga-MacKenzie et al Summix: A method for detecting and adjusting for population structure in genetic summary data. Am J Hum Genet. 2021 Jul 1;108(7):1270-1282. doi: 10.1016/j.ajhg.2021.05.016.

Summix relies on reference populations, but what if the ref pop is small?

1/

#STATGEN2024 #Genetics #StatisticalGenetics

STATGEN 2024 talk
Genotype prediction of 336,463 samples from public expression data
Afrooz Razi

recount3: uniformly processed RNA-seq
https://rna.recount.bio/

We developed a statistical model to predict genotypes from the Recount3 data

It has high prediction accuracy.

1/

#STATGEN2024 #Genetics #StatisticalGenetics

STATGEN 2024 talk
BRCAPRO+BCRAT: extending a Mendelian breast cancer risk prediction model to include non-genetic risk factors
Zoe Guan

BRCAPRO: Mendelian model, genes

BCRAT: 1st family hx, hormonal risk factors, hx of benign disease

Combine these complementary models.

https://www.mdpi.com/2072-6694/15/4/1090

#STATGEN2024 #Genetics #BreastCancer #RiskPrediction #StatisticalGenetics

STATGEN 2024 talk
Polygenic risk score analysis for multiethnic populations
Chris Amos

Polygenic Risk Scores (PRS)
* Inform re biological processes
* Identify some at higher risk
* Might motivate behavioral change

PRS could inform when to start screening.

"measles plot instead of a manhattan plot" - has excessive false positives all over the genome.

Lung cancer risk snp also is related to response to smoking cessation

1/

#STATGEN2024 #Polygenic #StatisticalGenetics #Genetics #PRS

STATGEN 2024 talk
Bayesian Meta-Analysis of Penetrance for Cancer Risk with Adjustment for Ascertainment Bias
Swati Biswas

Need accurate estimates of age-specific penetrance for cancer risk variants.
https://arxiv.org/abs/2304.01912

Heterogeneous studies w/ different measures of risk
Marabelli et al. Penetrance of ATM Gene Mutations in Breast Cancer: A Meta-Analysis of Different Measures of Risk. Genet Epidemiol. 2016 doi: 10.1002/gepi.21971

1/

#STATGEN2024 #Genetics #StatisticalGenetics #Pentrance

STATGEN 2024 talk
Improving Genetic Risk Prediction with Genetic Architecture and Functional Annotations
Wei Jiang

Genome-wide Empirical Bayes to use both genetic architecture and functional annotations in a computationally efficient way.

* Summary-statistics-based
* No parameter tuning needed
* Has improved prediction accuracy over existing methods.

https://www.researchsquare.com/article/rs-3266942/v1

#STATGEN2024 #Genetics #StatisticalGenetics

STATGEN 2024 talk
Novel Methods for Estimating Risk Parameters Associated with Polygenic Scores Using Case-Parent Trio Designs
Ziqiao Wang

Estimates of SNP effect sizes can be biased due to
* Population stratification
* Assortative mating

Prior method
PRS TDT (pTDT) (Weiner et al., Nat Genet 2017)

Goal
To develop a joint model that is flexible and robust

Assume family PGS ~ multivariate normal distribution w/ family-specfic mean & var

1/

#Genetics #StatisticalGenetics #STATGEN2024

STATGEN 2024 talk
Linking variants to gene networks with multivariate association approaches
Xuanyao Liu

Detecting trans-eQTLs is challenging
- Small trans- effects
- Multiple-testing correction
- Overwhelmed by false positives

Trans-PCO method

PCO = PC-based omnibus test

https://github.com/liliw-w/Trans

1/

#STATGEN2024 #StatisticalGenetics #Genetics

STATGEN 2024 talk
Localizing Rare-Variant Association Regions via Multiple Testing Embedded in an Aggregation Tree
Jichun Xie

Which variants
* Gene region
* Sliding window (fixed size)
* Varying window

DYNamic Aggregation TEsting (DYNATE) algorithm

"DYNATE dynamically and hierarchically aggregates smaller genomic regions into larger ones"

https://cran.r-project.org/package=DYNATE

1/

#STATGEN2024 #Genetics #StatisticalGenetics #RareVariants

STATGEN 2024 talk
Quantile regression GWAS with related samples
Fan Wang

Quantile regression tests whether a genetic variant associates with various quantiles of a trait.

Quantile Rank Score test
- Distribution-free
No transformation needed.
- Very fast
Estimate the null model only once.
- R package: QRank https://cran.r-project.org/package=QRank

1/

#STATGEN2024 #StatisticalGenetics #Genetics #QuantileRegression

STATGEN 2024 talk
Distinct explanations underlie gene-environment interactions in the UK Biobank.
Arun Durvasula

Genetic effects across the genome may exhibit context dependence
- European vs. East Asian genetic correlation is less than 1 across a wide range of traits.
- Hinting at polygenic GxE

GxE can arise through different scenarios
- Imperfect genetic correlation
- Varying genetic variance
- Proportional amplification

1/

#Genetics #StatisticalGenetics #STATGEN2024 #GxE

STATGEN 2024 talk
The influence of antipsychotic exposure on genetic susceptibility to obesity
Anne Justice

Many factors contribute to obesity risk, including medications.

Obesity Related to Antipsychotic Liability & Exposure (ORAcLE) Genetics Consortium
https://sites.wustl.edu/oracle/

Examine polygenic risk scores (PRS) for antipsychotic-induced weight gain in Geisinger MyCode, which began in 2007. 184,293 with genotype & whole exome data.

1/

#Genetics #STATGEN2024 #StatisticalGenetics #Obesity

STATGEN 2024 talk
Detecting latent systemic structure in deep phenotyping and genotyping data
Audrey Hendricks

Expecting systemic structure S to be the same/similar across all the traits.

Trait_i = X_i + E_i + (O_i + S)

How to infer S?

Multitrait finite mixture of regressions (MFMR) by Dahl et al (2019)

1/

#Genetics #StatisticalGenetics #STATGEN2024

STATGEN 2024 talk
Statistical Methods for Single-Cell RNA-Seq Analysis and Spatial Transcriptomics
Rafael Irizarry

tSNE and UMAP plots:
"They really aren't informative, but they are really pretty."

Negative control scRNAseq data set: the percent of zeros is very high, and contributes strongly to the first PCA. tSNE plot 'discovers' new cells.

Transformed to log2(1 + CPM): looks zero-inflated.

Raw counts: Poisson

1/

#Genetics #STATGEN2024 #StatisticalGenetics #RNAseq #Transcriptomics

STATGEN 2024 talk
Identifying GxE through Mendelian Randomization
Xiaofeng Zhu

Statistical power is low and detecting GxE is a challenge.

See:

Aschard H. A perspective on interaction effects in genetic association studies. Genet Epidemiol. 2016 Dec;40(8):678-688. doi: 10.1002/gepi.21989. Epub 2016 Jul 7. PMID: 27390122; PMCID: PMC5132101.

1/

#Genetics #STATGEN2024 #StatisticalGenetics #MendelianRandomization

STATGEN 2024 talk
An efficient method for network Mendelian randomization allows network structure discovery and effect estimation.
Jean Morrison

Existing methods

* GenomicSEM
* Network deconvolution
Graph-cML
bimmer

Our method

* Network empirical shrinkage Mendelian randomization (NESMR)
- likelihood-based

Assumptions
1. Causal effects between traits are linear with no interactions

1/

#Genetics #MendelianRandomization #STATGEN2024 #StatisticalGenetics

STATGEN 2024 talk
Synthetic Variables for Genetic Analysis with Censored Outcomes
Jin Zhou

ACCORD trail in T2D: more tightly controlling glycemic mean levels led to increase in deaths due to CVD.

But variation in glucose levels is a risk factor.

What factors influence glucose levels of variability?

Developed method to do a GWAS of trait variability within a longitudinal context.

Developed a fast robust estimating equations method

1/

#Genetics #StatisticalGenetics #Censoring #STATGEN2024

STATGEN 2024 talk
A New Test for Trait Mean and Variance Detects Unreported Loci for Blood Pressure Variation
Todd L. Edwards

If there is a GxE, then variance of Y differs by genotype.

If we don't know E, we can model both the mean & variance as function of the main predictor.

Known a long time - e.g. Waddington (1942) Canalization of development and the inheritance of acquired characters.

1/

#Genetics #STATGEN2024 #GxE #StatisticalGenetics

STATGEN2024 talk
Optimizing Polygenic Risk Scores for Diverse Populations
Nilanjan Chatterjee

Breast Cancer example:
Progress in developing PRS - OR per SD increasing from 1.49 (77 SNPs) to 1.64 (313 SNPs) to 1.71 (3,820 SNPs).

313 SNP PRS out-performed a 6-million SNP PRS. "Let's not get jazzed by the number of SNPs in the PRS".

Wanted to develop a method for better PRSs in diverse pops, so collaborated with 23andMe.

1/

#STATGEN2024 #Genetics #StatisticalGenetics

STATGEN 2024 talk
Cell type specific functional characterization of Alzheimer's disease in microglia
Yun Li

Yang, X., Wen, J., Yang, H. et al. Functional characterization of Alzheimer’s disease genetic variants in microglia. Nat Genet 55, 1735–1744 (2023). https://doi.org/10.1038/s41588-023-01506-8

AD SNP h^2 most highly enriched in microglia regulatory regions.

iPSC differentiation of microglia

Identified cis-regulatory elements (cCREs) near 37 AD loci

1/

#Genetics #StatisticalGenetics #STATGEN2024

STATGEN 2024
Interpreting structure in sequence count data with differential expression analysis allowing for grades of membership (GoM)
Peter Carbonetto

Allow a cell to be a partial member of > 1 group.

GoM is closely related to Non-negative Matrix Factorization (NMF).

Comparing k groups - a more stringent measure - drives more to zero than DESeq2 and has more power.

Groups = 'topics'. Can be cell-types or groups of cell-types, or more general

#Genetics #StatisticalGenetics #STATGEN2024

STATGEN 2024
Pleiotropy-robust methods for high-dimensional multivariable Mendelian randomization (HDMR)
Nathan LaPierre presenting, co-authors: Matthew Stephens, Xin He

In HDMR, we have many genetically correlated exposures, which may be explained by unobserved shared factors. These can be inferred by factor analysis.

Flexible, modular framework: Factor-Augmented MR
1. Factor Analysis
2. Regression/Variable Selection

#Genetics #StatisticalGenetics #MendelianRandomization #STATGEN2024

The keynote speaker opening the "STATGEN 2024: Conference on Statistics in Genomics and Genetics" is Kathryn Roeder, talking about "Testing of differential genomic outcomes in the presence of unmeasured confounding and missing data".

When testing gene expression across the genome, the majority of genes will follow the null. This enables QC checks, as the majority will not follow the null if we haven't adjusted adequately for unmeasured covariates.

#StatisticalGenetics #Genetics #STATGEN2024