This software package implements three well-calibrated statistical methods for analyzing cohorts of rare disease patients to find:
- genes recurrently impacted by de novo mutations across the cohort
- genes recurrently impacted by inherited compound heterozygous variants across the cohort
- genes harboring significant compound heterozygous variants in individual patients
Our RaMeDiES wiki also details how we ran our pathway analysis to find biologically-related groups of genes impacted with candidate variants across phenotypically similar patients.
If you use RaMeDiES in your work, please cite our publication:
SN Kobren*, MA Moldovan*, R Reimers, D Traviglia, X Li, D Barnum, A Veit, J Willett, M Berselli, W Ronchetti, R Sherwood, J Krier, IS Kohane, Undiagnosed Diseases Network, SR Sunyaev (2024). "Joint, multifaceted genomic analysis enables diagnosis of diverse, ultra-rare monogenic presentations." bioRxiv. doi: 10.1101/2024.02.13.580158.
- Python 3.6+
- Python libraries: os, sys, argparse v1.1+, numpy v1.23.3+, scipy v1.91+
- ❗ Operating System: Linux distribution; compatibility on MacOS is not guaranteed, and Windows is not supported.
Edit the configuration cfg.py file to include the full path to your local installation of this repository.
script_directory = "/full/path/to/github/repo/RaMeDiES/"
All RaMeDiES statistical models operate at the level of mutational targets, which intuitively correspond to the total mutation rate of all possible variants (of a particular type) within a gene. We have precomputed per-gene mutational targets for CADD and SpliceAI variant functionality scores with respect to GRCh38/hg38.
You must download these seven required files from Harvard Dataverse and store them locally in /full/path/to/github/repo/RaMeDiES/data:
ens2gene.txt.gz(136 KB)pseudogenes.txt.gz(231 KB)score_lists_CI.txt.gz(629 KB)score_lists_CS.txt.gz(97 MB)score_lists_II.txt.gz(218 KB)score_lists_IS.txt.gz(13.35 MB)shet_table.txt.gz(489 KB)
Descriptions of, sample code for running, and customizable parameters for the following steps of our statistical framework are detailed in our wiki:
- Preprocess input variant data
- Cohort-level de novo recurrence
- Cohort-level compound heterozygosity
- Individual-level compound heterozygosity
- Gene set enrichment across patient subgroups
❗ Our statistical models operate at the level of "mutational targets" rather than individual-level variant data. These intermediate computed files can be shared freely to enable cross-cohort meta-analyses! See our enabling cross-cohort analyses wiki page for more details.
If you have questions or comments about running any of the code found in this repository, please contact Shilpa Kobren or Mikhail Moldovan at [first name]_[last name] at hms.harvard.edu.