RaMeDiES: Rare Mendelian Disorder Enrichment Statistics

This software package implements three well-calibrated statistical methods for analyzing cohorts of rare disease patients to find:

1. genes recurrently impacted by de novo mutations across the cohort
2. genes recurrently impacted by inherited compound heterozygous variants across the cohort
3. genes harboring significant compound heterozygous variants in individual patients

Our RaMeDiES wiki also details how we ran our pathway analysis to find pathways enriched with candidate diagnostic 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, RI Corona, GdV Carvalho Neto, J Willett, M Berselli, W Ronchetti, SF Nelson, JA Martinez-Agosto, 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.

✨ Prerequisites

• Python 3.6+, R 4.1+
• Python libraries: os, sys, argparse v1.1+, numpy v1.23.3+, scipy v1.91+, rpy2 v3.15.16+, requests v3.31+, urllib3 v1.26.8+
• R packages: cluster
• ❗ Operating System: Linux or MacOS; Windows is not supported.

✨ Configuration

Edit the configuration cfg.py file to specify the name of the column in your input variant files corresponding to the deleteriousness score you'd like to use for coding SNVs:

vcf_format_dict = {"coding_snv_score" : "CADD-raw"}  # default is CADD-raw

Note that CADD is the only currently supported score for coding indels, and SpliceAI is the only currently supported score for intronic SNVs and indels.

✨ Precomputed data files

We have precomputed per-gene mutational targets for various variant functionality scores with respect to GRCh38/hg38. Pointers to the most up-to-date versions of these files can be found in /full/path/to/github/repo/RaMeDiES/data.

You will need to have the git Large File Storage (lfs) extension installed to download these files. It is supported on Mac, Windows, and Linux. On a Linux server without root access, you can install using:

wget -O git-lfs.tar.gz https://github.com/git-lfs/git-lfs/releases/download/v3.5.1/git-lfs-linux-amd64-v3.5.1.tar.gz
tar -xvzf git-lfs.tar.gz
mv git-lfs/git-lfs ~/bin
git lfs install --skip-repo

Then you can run git pull to download the following files, or, if that fails, git reset --hard instead:

• ens2gene.txt.gz (133 KB)
• pseudogenes.txt.gz (225 KB)
• gene_constraint_scores.txt.gz (245 KB)
• CADD_CI.txt.gz (614 KB)
• CADD_CS.txt.gz (93 MB)
• SpliceAI_II.txt.gz (213 KB)
• SpliceAI_IS.txt.gz (13 MB)
• AlphaMissense_MS.txt.gz (68 MB)
• PAI3D_MS.txt.gz (63 MB)
• REVEL_MS.txt.gz (67 MB)

✨ RaMeDiES Framework

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.

✨ Contact

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.