giab

(current) A NGS analysis framework for WGS data, which automates the entire process of spinning up AWS EC2 spot instances and processing FASTQ to snvVCF in <60m, for dollars a sample and achieving Fscores of 0.998.

Nextflow DSL2 pipeline for human WGS (GRCh38, ~30x): structural variants (Manta + Delly), CNVs (CNVpytor + GATK gCNV), STR expansions, SMN1/SMN2 copy number, alpha-globin dosage and copy-number traits. Docker-based, with a per-sample HTML report. Research and educational use only.

GPU-accelerated germline WGS pipeline for the HG002 reference sample, built with NVIDIA Parabricks

13-process Nextflow DSL2 germline WES workflow from FASTQ to PASS variants, validated on HG002; separate GIAB benchmarking, VEP annotation, and exploratory variant prioritization.

Snakemake workflow that calls germline small variants with bcftools and FreeBayes and benchmarks them against GIAB HG002 with RTG vcfeval, by difficult-region stratum.

PyTorch implementation of Symmetric σ-AS-RectLU and Smooth RectLU for severe class-imbalanced CNNs. Includes shortcut-controlled NIST GIAB 1D/2D benchmarks, Clair3/DeepVariant-style pileup backbones, and automated unit tests.

Production Nextflow DSL2 germline variant calling pipeline with biomarker reproducibility analysis and AWS execution

ML pipeline for structural variant detection from ONT/PacBio long-read sequencing: Minimap2 + Sniffles2/cuteSV candidates, Random Forest/XGBoost classification, GIAB HG002 benchmarking, SHAP explainability. Snakemake-orchestrated, reproducible.

Provenance-first sequencing benchmark evidence registry and Evidence Atlas for inspectable, reproducible cross-platform evidence.

Pipeline Nextflow DSL2 de variant calling germinal sur chr20, benchmarke contre le truth set GIAB HG002 avec hap.py. GATK, bcftools et DeepVariant compares sur donnees identiques.

Variant calling benchmarking on GIAB HG002 (WES chr22) using hap.py; SNP/INDEL Precision/Recall/F1.

Reproducible benchmarking of DeepVariant and GATK callsets against GIAB HG002, with gene-level metrics and tandem-repeat context analysis.

Nextflow DSL2 germline SNV/indel calling (BWA-MEM, GATK HaplotypeCaller, bcftools csq, ClinVar) benchmarked on GIAB HG001: SNV F1 0.996

Release validation for a germline WGS workflow (WDL): GATK and DeepVariant benchmarked against GIAB HG002, with a regression gate, case studies and runbooks

Reproducible Snakemake pipeline for germline variant calling (BWA-MEM + bcftools), benchmarked against the GIAB truth set with RTG vcfeval

GATK4 germline variant calling (Nextflow DSL2): BWA-MEM -> MarkDuplicates -> BQSR -> HaplotypeCaller -> hard-filter, benchmarked against a GIAB truth set (precision/recall/F1).

This repository contains data indexes from NIST's Genome in a Bottle project.

Benchmark asking whether a variant caller's confidence is honest, not just accurate — GIAB-scored performance stratified by genomic difficulty, QUAL calibration, and quality filtering treated as an abstention decision

Publicly available statistical genetics data and analysis. 1000G, HGDP, SGDP, and GIAB. Reproducible global ADMIXTURE with variable K.

Validation and change control of a germline NGS panel pipeline (BWA-MEM, GATK) against GIAB reference samples, documented the way a diagnostic laboratory would