Genome-Wide Association Studies using the ADAM genomics analysis platform. This is work-in-progress. Currently, we support Additive Linear and Dominant Linear models (for continuous outcome variables), and Additive Logistic and Dominant Logistic models (for binary [0,1] or [1,2] outcome variables). Each can be run with or without phenotypic covariates.
To build, install Maven. Then (once in the gnocchi directory) run:
mvn package
Maven will automatically pull down and install all of the necessary dependencies.
Occasionally, building in Maven will fail due to memory issues. You can work around this
by setting the MAVEN_OPTS environment variable to -Xmx2g -XX:MaxPermSize=1g.
To run, you'll need to install Spark. If you are just evaluating locally, you can use a prebuilt Spark distribution. If you'd like to use a cluster, refer to Spark's cluster overview.
Once Spark is installed, set the environment variable SPARK_HOME to point to the Spark
installation root directory. Then, you can run gnocchi via ./bin/gnocchi-submit.
We include test data. You can run with the test data by running:
./bin/gnocchi-submit regressPhenotypes testData/5snps10samples.vcf testData/10samples5Phenotypes2covars.txt ADDITIVE_LINEAR testData/associations -saveAsText -phenoName pheno1 -covar -covarFile testData/10samples5Phenotypes2covars.txt -covarNames pheno4,pheno5
We accept phenotype and covariate inputs in CSV (tab delimited) format:
SampleID pheno1 pheno2
mySample pheno1_value pheno2_value
Note: phenotypes and covariates must be numerical. For nominal scale data (i.e. categorical data), binarize. For ordinal scale data, convert to integers.
This project is released under an Apache 2.0 license.