PheMED

Phenotypic Measurement of Effective Dilution

Dependencies

After installing Anaconda or Miniconda, running the following commands will create and activate an environment suitable to run PheMED.

conda env create --file environment.yml
source activate phemed

Running PheMED

To run PheMED on the sample data, run the following command:

python phemed.py --sum_stats data/sim_data.csv --n_studies 2 --out output/local_test

• --sum_stats identifies the csv with the merged log odds ratios and standard errors for each study
• --n_studies denotes the number of studies being analyzed
• --out denotes prefix for output files

For the sum_stats csv, PheMED expects the input csv file to have columns: SNP, CHR, and POS corresponding to the rsid, chromosome and base pair position of the SNP. The columns that follow are assumed to contain effect sizes {log(OR)₁,log(OR)₂,...,log(OR)_n} followed by the SEs {SE₁,SE₂,...,SE_n}, when analyzing n studies. Furthermore, the first study listed is used as the reference study when measuring effective dilution. (See the sim_data.csv in the data directory for 2 studies or an example with three studies below.)

SNP	CHR	POS	STUDY1	STUDY2	STUDY3	SE1	SE2	SE3
rs1	1	1	0.069	0.010	0.040	0.044	0.01	0.027
rs2	1	2	0.038	0.015	0.026	0.044	0.01	0.027
rs3	1	3	0.045	-0.0004	0.022	0.044	0.01	0.027

Currently, the phemed.py script assumes that study samples do not overlap. For guidance on running PheMED when studies have sample overlap, please see the FAQ.

Understanding Outputs: PheMED produces both a log file and a csv <out>_Summary.csv with the relevant outputs. The csv file includes the effective dilution values (in the PheMED column). If CI's are computed, users can find the bounds of the bootstrapped 95% confidence interval with the columns CI_.025 and CI_.975. Furthermore, if CI's are computed, <out>_Summary.csv will also contain p-values for the effective dilution, denoted by the column P (see section P Values below for details). Finally, if the user provides sample size information for each of the studies, the csv will also produce a column estimating the dilution adjusted effective sample size in the DilutionAdjNEff column. For examples of output, see the output directory.

P Values: For computing the p-value we leverage three different p-value methodologies. For each of the methodologies, there is a corresponding qc column that indicates if it is appropriate to use that methodology to compute p-value. As such, we do not require that all three methodologies pass the QC check; instead, only one such methodology needs to pass the QC check to estimate the p-value.

Nevertheless, for naive count, if the p-value does not pass the QC check (e.g. the p-value is very small and becomes hard to estimate from the bootstrap simulation), if the number of bootstrap samples is sufficiently large (e.g. = 2000, the default), we can still use the confidence intervals to infer if p < .05. See our paper for details.

Brief Overview of Other Arguments Computing the CIs when measuring the effective dilution between two GWA studies can take around an hour. If you would like to run the pipeline without computing the CIs, use --compute_cis False or the --n_CIs <smaller integer> to reduce the number of bootstrap samples; however decreasing the number of samples below the default value can result in noisy p-values. While sample sizes are not required to run PheMED, if you would like to compute a dilution adjusted effective sample size, you will need a csv indicating the sample sizes for each study maintained in the same order as the input file and add either the argument --eff_sample_sizes <path_to_file> or --sample_sizes <path_to_file>, where the former argument expects a CSV with a column labeled N detailing the effective sample sizes and the later argument expects a CSV with columns labeled Cases and Controls, that detail the number of cases and controls for each study. For examples of these files, see data/eff_sample_size_sim.csv and data/sample_size_sim.csv respectively.

For additional arguments run

python phemed.py -h

Performing Dilution Adjusted Weights (DAW) Meta-Analysis with PheMED

After running the phemed script above, users can run a dilution adjusted weights meta-analysis with the following command

python daw_meta.py --sum_stats data/sim_data.csv \
                  --n_studies 2 \
                  --out output/local_test \
                  --dilution_weights output/local_test_DilutionVals.csv

• --sum_stats identifies the csv with the merged log odds ratios and standard errors for each study
• --n_studies denotes the number of studies being analyzed
• --out denotes the prefix for output files
• --dilution_weights path to the summary csv outputted from the phemed script

The script will then output a csv containing meta-analyzed Z-scores, p-values and effect sizes as measured according to the reference study. Example output below (for the 3 studies input):

SNP	CHR	POS	Z_META	BETA_META	P_META
rs1	1	1	2.216573171014446	0.04237889772244553	0.026652272616020767
rs2	1	2	2.001733259609271	0.03827135065330354	0.04531342729038265
rs3	1	3	0.8086750926075479	0.015461144927879799	0.41870205861603993

FAQ: For FAQ and troubleshooting, please see the FAQ here.

Support: If you have any outstanding questions not addressed by the docs or FAQ, feel free to post your question as a GitHub Issue here or send an email to david.burstein2 {at} mssm.edu

Citing our Paper: If you use our software, please cite our preprint on medRxiv: Burstein et al. Detecting and Adjusting for Hidden Biases due to Phenotype Misclassification in Genome-Wide Association Studies