This repository contains the necessary scripts to build a docker container for executing VARPP-RuleFit and to predict with the same model on a patient .vcf file.
Docker needs to be installed on the machine you are pulling the image to.
To install the image it needs to be pulled from DockerHub in the following manner
docker pull greenlantern/varpp-predict-utilsPlease note that this image is only one of several components necessary to run the model. The image is enough to create a custom VARPP-Rule model, but requires extra data to predict with it.
The docker image is part of a DataLad setup, that requires:
- positional data files for genetic variants across all genes available in the GTEx and HCL data sets
- In case of prediction with pre-computed models, the
VARPP-Ruleresults files for all HPO terms tested in the associated publication - The container to prepare the patient variant file, in case it is annotated with
hg19instead ofhg38reference genome information.
To get help on how to create a custom VARPP-RuleFit model to be used for prediction of pathogenic variants of a patient, run the following:
docker run greenlantern/varpp-predict-utils varppRule.R --help Usage: /root/varpp-prediction/varppRule.R [options] VARPP-RuleFit. Options: -g CHARACTER, --gene_list=CHARACTER A list of potentially pathogenic genes associated with the patient. -t CHARACTER, --type=CHARACTER Run VARPP-RuleFit with either 'hcl', 'gtex' or 'custom' data -p CHARACTER, --user_pathogenic=CHARACTER A file with custom pathogenic data in the same format as the HCL or GTEx data. -b CHARACTER, --user_benign=CHARACTER A file with custom benign data in the same format as the HCL or GTEx data. -o CHARACTER, --output_path=CHARACTER Specify the path to the results. Defaults to the 'results' in the current directory. -c CHARACTER, --cores=CHARACTER Maximum number of cores available. This will be distributed across the model -n CHARACTER, --ntree=CHARACTER Number of trees in the model, defaults to 2000. -h, --help Show this help message and exit Example: ./varppRule.R -g <gene list> -t <type: GTEx, HCL, custom> -p <User provided pathogenic data> -b <User provided benign data> -o <output/path> -c <Maximum number of cores available> -n <Number of trees, defaults to 2000>
This indicates all the options required for running the program.
In order to predict with the model we created in the `Run the model` step, use the predict.R function of the container:
docker run greenlantern/varpp-predict-utils predict.R --help varpp prediction script. Options: -v CHARACTER, --patient_vcf=CHARACTER A patient derived .vcf file. -d CHARACTER, --data=CHARACTER Model data to predict with. Either GTEx or HCL. -p CHARACTER, --hpo=CHARACTER HPO term ID to predict with. This needs to be in the list of HPO terms that were pre-modelled. -c CHARACTER, --custom_model=CHARACTER Path to a pre-trained VARPP-RuleFit model in the form of a .rds object If this option is chosen,-p/--hpo option will be ignored. -o CHARACTER, --output_path=CHARACTER Specify the path to the results. Defaults to the 'results' in the current directory. -h, --help Show this help message and exit Example: ./predict.R -v <patient.vcf> -d <data: GTEx or HCL> -p <HPO term ID> -o <output/path>
The most complete way of running the VARPP-RuleFit model is to use the varppRuleAndPredict.R function:
docker run greenlantern/varpp-predict-utils varppRuleAndPredict.R --help Usage: /root/varpp-prediction/varppRuleAndPredict.R [options] VARPP-RuleFit. Options: -v CHARACTER, --patient_vcf=CHARACTER A patient derived .vcf file. -g CHARACTER, --gene_list=CHARACTER A list of potentially pathogenic genes associated with the patient. -t CHARACTER, --type=CHARACTER Run VARPP-RuleFit with either 'hcl', 'gtex' or 'custom' data -p CHARACTER, --user_pathogenic=CHARACTER A file with custom pathogenic data in the same format as the HCL or GTEx data. -b CHARACTER, --user_benign=CHARACTER A file with custom benign data in the same format as the HCL or GTEx data. -o CHARACTER, --output_path=CHARACTER Specify the path to the results. Defaults to the 'results' in the current directory. -c CHARACTER, --cores=CHARACTER Maximum number of cores available. This will be distributed across the model -n CHARACTER, --ntree=CHARACTER Number of trees in the model, defaults to 2000. -l CHARACTER, --lasso=CHARACTER Number of LASSO bootstrap rounds. -m CHARACTER, --max_depth=CHARACTER Maximum tree depth, defaults to 3. -h, --help Show this help message and exit Example: ./varppRuleAndPredict.R -v <path to patient vcf file> -g <gene list> -t <type: GTEx, HCL, custom> -p <User provided pathogenic data> -b <User provided benign data> -o <output/path> -c <Maximum number of cores available> -n <Number of trees, defaults to 2000> -l <Number of LASSO bootstrap rounds> -m <Tree depth>
docker run -v $(pwd):/mnt greenlantern/varpp-predict-utils:latest varppRuleAndPredict.R \
-v patientVcf/patient.vcf \
-g ASL,CLCNKB,KCNJ10,SLC12A3,SCNN1B,SCNN1G,HSD11B2,AIP,CLCNKA,BSND,SLC12A1,KCNJ1,SARS2,OTC,NR3C1,CACNA1D,ASS1,CA5A,SLC26A3,SLC25A15,USP8,NARS2,IKZF1,CYP17A1,KCNJ5,HLA-B,SCNN1A,CPS1,NOS1,REN \
-t gtex \
-c 64 \
-l 100 \
-m 3The console will show an output similar to the following:
Predictions are: Chr Start End Gene CADD_raw_rankscore Prediction 1 chr1 941187 941188 SAMD11 22.200 0.9999974 2 chr1 955212 955213 NOC2L 0.267 0.9999916 3 chr1 968669 968670 PLEKHN1 1.024 0.9999995 4 chr1 1006171 1006172 ISG15 3.296 0.9999857 5 chr1 1023443 1023444 AGRN 1.481 0.9999972 6 chr1 1024497 1024498 AGRN 1.584 0.9999973
and the results are stored in scratch in the current working directory as:
predictions.csv VARPP-RuleFit_results.rds