ClinicalVariantR — Variant Prediction Platform

Version: 0.99.3 (Bioconductor x.99.y devel)

ClinicalVariantR is an R Shiny application for ACMG/AMP evidence-based germline variant prediction with per-variant curator workflow.

Repository: https://github.com/safarafique/ClinicalVariantR

Bioconductor submission docs

Package metadata and submission guides live under inst/Bioconductor/:

Document	Purpose
inst/Bioconductor/SUBMISSION_CHECKLIST.md	Required files + readiness status
inst/Bioconductor/PACKAGE_CONVERSION.md	Convert Shiny-app layout → Bioconductor package
inst/Bioconductor/ISSUE_TEMPLATE_DRAFT.md	Draft for BiocContributions issue

Core package files: DESCRIPTION (version 0.99.3), NAMESPACE, LICENSE, NEWS.md, man/, vignettes/ClinicalVariantR.Rmd, tests/, inst/CITATION.

Install and launch (Bioconductor Shiny style — return app, user runs it). On Windows, restart R first if the package is already loaded:

# install.packages("remotes")
# Preferred while developing from this clone:
remotes::install_local("E:/ACGM/ClinicalVariantR", force = TRUE, upgrade = "never")
# Or from GitHub after the latest launcher is pushed:
# remotes::install_github("safarafique/ClinicalVariantR")
library(ClinicalVariantR)
app <- ClinicalVariantR()
if (interactive()) shiny::runApp(app, launch.browser = TRUE)

# Fast path without reinstall (avoids package R/ auto-load warning):
# shiny::runApp("E:/ACGM/ClinicalVariantR/inst/shinyapp", launch.browser = TRUE)
# Or from package root (app.R delegates to inst/shinyapp):
# shiny::runApp("E:/ACGM/ClinicalVariantR", launch.browser = TRUE)

On Windows with R 4.6, install Rtools 4.5+ before devtools::check() / package builds. Rtools 4.4 is not compatible. Until Rtools is fixed, you can still check without compiling:

devtools::check(vignettes = FALSE, args = "--no-manual", error_on = "never")

Important: documentation scaffold is ready; full package conversion (paths to inst/, remove source() chains, clean R CMD check / BiocCheck) must finish before opening a BiocContributions issue.

Pipelines

Pipeline	Inputs	Scope
Group A (Clinical prediction)	VCF + Clinical CSV + Pedigree CSV	28 criteria + per-variant curation
Group B (Automated prediction)	VCF only	18 automated criteria
Group C (Gene panel)	VCF + gene list	Group B engine, panel-filtered

Prediction mode (v2.5+): HPO/OMIM phenotype matching (PP4), trio genotype parsing (PS2/PM3/BP2), ClinGen AutoPVS1, login + encrypted uploads, formal validation report.

Prediction mode (v2.4+): stricter PVS1 (LoF panel required), evidence strength labels, prediction limitations per variant, reference readiness checks.

See config/prediction_reference_install.md for full gnomAD/ClinVar/REVEL setup.

Authentication and PHI storage (v2.5)

• Authentication is disabled for local development by default (AUTH_ENABLED <- FALSE in R/auth_storage.R). For secured demos, provide users through config/auth_users.csv or the CLINICALVARIANTR_USER / CLINICALVARIANTR_PASSWORD environment variables.
• Override with environment variables CLINICALVARIANTR_USER and CLINICALVARIANTR_PASSWORD when no user file is present.
• Uploads are encrypted at rest (AES-256-GCM via OpenSSL) under logs/secure_uploads/; set CLINICALVARIANTR_ENCRYPTION_KEY (32 chars) for production.
• Access events are logged to logs/access_audit.csv.

Validation report

cd ClinicalVariantR
Rscript scripts/generate_validation_report.R

Expert-classified variants live in data/validation/clinicalvariantr_validation_set.tsv (expand toward 500+ for production sign-off).

Features include VCF requirement validation (green/red readiness), variant preview, complete large-VCF streaming analysis (no row cap), optional bcftools integration on Ubuntu/WSL, color-coded results, audit logging, and CSV export.

Quick start

Install and open the app

First launch from this clone auto-installs missing R packages (CRAN + Bioconductor), then opens the app:

shiny::runApp("E:/ACGM/ClinicalVariantR", launch.browser = TRUE)

Or install dependencies once, then launch:

# Optional one-shot install (same as first-run auto-install):
# Rscript scripts/install_app_deps.R
shiny::runApp("E:/ACGM/ClinicalVariantR", launch.browser = TRUE)

Package install path:

install.packages("remotes")
remotes::install_local("E:/ACGM/ClinicalVariantR", force = TRUE, upgrade = "never", dependencies = TRUE)
library(ClinicalVariantR)
app <- ClinicalVariantR()
if (interactive()) shiny::runApp(app, launch.browser = TRUE)

To install dependencies once from a clone (not used by the Bioconductor package tarball):

# Rscript scripts/install_app_deps.R
# Or: source("scripts/bootstrap_deps.R"); clinicalvariantr_ensure_dependencies()

Run from a local clone (no reinstall)

shiny::runApp("E:/ACGM/ClinicalVariantR", launch.browser = TRUE)

Ubuntu / WSL (recommended for large VEP VCFs)

cd /mnt/e/ACGM/ClinicalVariantR
bash scripts/ubuntu_setup.sh          # bcftools + R CLI packages
# Or install R CLI deps only:
Rscript scripts/install_r_cli_deps.R  # data.table, readr, jsonlite → ~/R/.../library
R -e "shiny::runApp('.', host='0.0.0.0', port=3838)"

Verify scripts (after install_r_cli_deps.R completes):

Rscript scripts/verify_group_a_28.R
Rscript scripts/verify_group_b_c.R
Rscript scripts/generate_validation_report.R

Expert review (after prediction)

Each pipeline includes an Expert review checklist accordion with worklist counts and filtered CSV exports:

• Export LP+ only — Pathogenic and Likely Pathogenic variants (priority review).
• Export expert worklist — LP+ plus VUS with ≥2 pathogenic evidence; excludes PM2-only VUS.

Use the full prediction report for audit; use the worklist export to start manual sign-out review.

Project structure

For a fuller file-by-file orientation, see CODEBASE_MAP.md.

ClinicalVariantR/
├── app.R
├── global.R
├── ui.R
├── server.R
├── R/                  # ACMG logic, VCF parsing, validation, streaming
├── data/reference/     # gnomAD, ClinVar, REVEL placeholders
├── data/samples/       # Example VCF and CSV files
├── scripts/            # ubuntu_setup.sh, validate_logic.R
├── logs/               # Runtime audit logs (gitignored)
└── www/                # Custom CSS

Complete VCF analysis options

Option	Default	Purpose
Analyze entire VCF	ON	Process every row (no limit)
FILTER=PASS only	OFF	Optional QC pre-filter
Minimum QUAL	0	Drop low-quality rows
Use bcftools	auto	Ubuntu/WSL speed boost
Chunk size	10,000	Variants per memory batch

Outputs

Best input for expert review: download the ClinicalVariantR CSV export after analysis completes.

Export	When to use
ClinicalVariantR_Expert_Worklist_*.csv	Start here — LP+ and high-priority VUS for sign-out review
ClinicalVariantR_LP_Plus_*.csv	Pathogenic / Likely Pathogenic only
ClinicalVariantR_Prediction_Report_*.csv	Full audit trail — all variants and evidence

Key columns for expert review: variant_id, gene, classification, criteria_met, criteria_rationale, prediction_limitations, evidence_json, confidence_score, evidence_strength.

Also written automatically under logs/ during streaming runs. Audit trail: logs/analysis_log.csv.

Logic check

setwd("path/to/ClinicalVariantR")
source("global.R")
test_acmg_logic_engine()

License

MIT (add LICENSE file if required for Bioconductor submission)