Quantifying the combined heritability of a trait based on a multi-ethnic LD panel with equal distribution of samples among each ancestry group.
Heritability of a trait is often identified and reported in an ancestry group stratified manner. This limits the ability to estimate and report the combined heritability in a multi-ethnic population. Although there are several methods demonstrated recently with robust ways of calculating heritability with or without individual-level datasets, these methods are limited to ancestry-specific groups. In this project, we are proposing a way to calculate combined heritability using a multi-ethnic reference linkage-disequilibrium (LD) panel with equal proportions of data. We will use current existing tools to simulate and calculate heritability and report it as a framework that can be implemented and explored further. This will lead to the development of a novel approach to estimating the heritability of particular traits in multi-ethnic populations. As a part of Team HeriVar, you will be contributing to the demonstration of methodology, calculation of heritability, and work as a team to promote the method.
With the increasing availability of multi-ethnic whole genome sequence datasets, there is a gaping absence of an approach to estimate the heritability of particular phenotypic trait that accounts for the multi-ethnic genetic architecture. This approach of calculating the combined multi-ethnic heritability has not been pursued previously. This project helps us understand the problems facing this issue in the field of genomics and helps in generating a framework using existing tools to calculate and assess the heritability of a trait in multi-ethnic populations.
- High Coverage 1000g dataset downloaded from http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1000G_2504_high_coverage/working/20201028_3202_phased/
- GWAS summary statisitcs for NTproBNP (In house) & BP downloaded from Pan-Ukbiobank analysis. (https://pan.ukbb.broadinstitute.org/phenotypes)
- R. ( module load R )
- Python. ( module load Anaconda3 )
- PLINK (https://www.cog-genomics.org/plink/2.0/ or module load PLINK in Cheaha).
- LDAK/SUMHER (https://dougspeed.com/sumher/).
- LDSC (https://github.com/bulik/ldsc).
- LiftOver ( https://genome.ucsc.edu/cgi-bin/hgLiftOver )
- LDSC requires Anaconda3 or Python-2.7 and subpackages like bitarray, nose, pybedtools, scipy, numpy, pandas, bioconda. (will be installed when generating environment).
- SumHer uses Intel MKL Libraries as dependencies. ( module load imkl/2020.1.217-iimpi-2020a )
-
LDSC ( Required to be installed by everyone in their home directory to use it )
- Clone the github of ldsc (git clone https://github.com/bulik/ldsc.git) and cd into the folder
- Module load Anaconda3 ( module load Anaconda3 )
- Install dependencies using conda as suggested by github ( conda env create --file environment.yml )
- Activate ldsc ( source activate ldsc )
- Test installation by running python scripts shared as path of repo ( ./ldsc.py -h )
-
Sumher
- Download the LDAK Linux executable file by requesting using name and email ( you will get an email from the developer with downloadables if you are a first time user )
- Unzip the executable file and use it. ( /data/project/ubrite/hackathon2022/staging_area_teams/HeriVar/Tools/ldak5.2.linux - It can be accessible by everyone)
- It also have executable for MAC users. Note: Please check Dependencies before installing the tools.
-
LiftOver
- Download the file from https://genome.ucsc.edu/cgi-bin/hgLiftOver
- Download the chain file needed for conversion - we can download it from above link.
- Run liftOver -h
- Datasets
-
We downloaded 1000g high coverage reference dataset from http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1000G_2504_high_coverage/working/20201028_3202_phased/.
-
We then extracted individuals files and randomly chose 489 unrelated individuals among each ancestry group.
-
Rationale behind including sample individuals from multiple ancestry groups is by taking equal number of individuals, we can have equal ld pattern distribution among the individuals.
-
Admixed population were excluded from the analysis along with related individuals which to 1956 individuals.
-
We removed variants with less than 1% minor allele frequency and variants with more than 5% missing data.
Allele Frequency Distribution among each ancestry and overall.
-
- PCA Analysis
-
We used Plink to calculate principal compnents analysis to test whether we have equal distributions of samples per ancestry group.
PC distributions stratified by Ancestry
-
- Prunning & Thresholding
- After subsetting to sample of interest, we did prunning and thresholding based on different cutoffs.
- Plink is used to generate the files needed.
- We used R2 and window size parameters for analysis.
-
R-squared cutoff of 0.2, 0.4, 0.6, 0.8.
-
Window size of 250kb, 500kb, 1Mb, 10Mb.
Distribution of Variants after P + T
-
-
We had ran near 1000 jobs for generating this datasets in Cheaha.
-
We decided to exclude High LD regions as recommended by the tools.
-
We subsetted the datasets to two categories.
- Pre HighLD regions removal.
- Post Hight LD regions removal.
-
Refernces panel generation
-
We used the two categories as mentioend above and used two tools to calculated reference LD panels.
-
We used ldsc to generate LD scores for all the categories we have.
LD_scores Distribution for Chromosome 22
-
-
For LDAK annotations, We used liftover to convert blk annotations from grch37 to grch38 and working on generting tagging files
- We had an issue generating LDAK annotations files and decided to pursue analysis after hackathon.
-
Phenotypes Processing
- We have also worked on processing phenotypes based as suggested by the tools.
-
Heritability
- We tried to generate h2 values using LDAK & LDSC but couldnt able to complete because of last minute issues.
- Akhil Pampana | pampana.akhil@gmail.com | apampana@uabmc.edu | Team Leader
- Nick Sumpter | nicks95@uab.edu | Team Member
- Yongyu (Frank) Qiang | frankqiang5040@gmail.com | Team Member