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.

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

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

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

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.

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

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.

Stratified variant-calling evaluation on GIAB chr20. Aggregate F1 hides 35x of the effect; skipping confident-region restriction penalises the better caller 4.5x.

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.

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

End-to-end germline variant-calling platform: Nextflow DSL2 pipeline benchmarked vs GIAB (SNV F1 0.9914), insert-only provenance, AWS CDK + Azure IaC, dbt/Airflow/Metabase, and guardrailed LLM variant interpretation.