First real app

WGSExtract made GUI for CLI prove itself.

The WGSExtract bundle is the first real application built through GUI for CLI. It pushed the project beyond demo controls into setup, command execution, data state, localization, and packaged releases.

Architecture Documentation

Why this bundle mattered

WGSExtract gave GUI for CLI a real target: a genomics workflow with long-running tools, user-selected files, setup state, reference data, and packaging expectations.

Setup

More than a launch button

The bundle needs setup steps, environment checks, external scripts, tool paths, and recoverable state before actions are safe to run.

Commands

Long-running process UX

Genome commands need terminal output, cancellation, exit code handling, path arguments, generated command previews, and clear disabled states.

Data

Dynamic rows and config

WGSExtract exercises data sources, state-dependent actions, persisted settings, localization, icons, and workspace files.

What v0.6.6 exposes

The desktop app bundles WGSExtract CLI v0.6.6. This update adds no new app controls: v0.6.6 reaches you through the actions already in the app, with no settings to change.

Oxford Nanopore in existing actions

Nanopore (ONT) reads are detected from PL:ONT read groups or UUID read names.

Align and Realign: use minimap2's map-ont preset and keep MM/ML methylation tags through alignment and sorting. Realign refuses its keep-FASTQ option for ONT input rather than lose those tags. FASTQs renamed without platform metadata, such as SRA downloads, are treated as Illumina, and the app has no platform override.
DeepVariant: selects the ONT_R104 model. It supports R10.4.1 chemistry only and needs x86-64, so ONT calling is refused on Apple Silicon and ARM64. DeepVariant itself is still a separate install.
SV Call: runs Sniffles2 instead of Delly. App setup does not install Sniffles; install it separately before running this action on ONT data.

Other improvements in existing actions

DeepVariant: CRAM 3.1 input is converted to a temporary CRAM 3.0 copy, and whole-genome postprocessing memory is estimated and capped.
Realign: paired reads stream through collation at whole-genome scale, and single-end input is refused before extraction instead of being dropped.
BAM Info: sex is inferred when chrX or chrY is missing, and depth binning is faster.
Microarray: coordinate projections and catalogues are cached for faster runs.

CLI only for now

v0.6.6 also adds Clair3 calling, CpG methylation pileups, Y-STR panels, queryable variant tables (TSV/SQLite/Parquet), report-issue, vendor chip detection, and JSON result manifests. None of these has an app action yet, so run them from the CLI. Native Linux ARM64 DeepVariant is CLI only too, because Linux ARM64 desktop installers are not published.

Controls from earlier releases

These came with GUI #171. They start unchecked or blank, inheriting CLI/config defaults rather than forcing them off.

FASTQ: aggregate existing FastQC reports into offline MultiQC HTML, with file and directory pickers and explicit report replacement. MultiQC must already be available; this action neither installs it nor runs FastQC.
Extract and Library: optional Poz Y-BAM regions (GRCh37/38), unmapped FASTQs, and allele-annotated yBrowse Y-SNP calling on matching indexed GRCh38. Install the pinned yBrowse resource explicitly from Library first.
Microarray: marker-only MyHeritage GSA and Ancestry layouts; AADR 1240K, HO, union, and opt-in v66.p1 2M PLINK/EIGENSTRAT research exports; quality/depth filtering, strand masking, historical rsIDs, no-call/reason reports, and a sequence-defined I/D manifest bound to one selected consumer format.
BAM Info: streaming per-chromosome depth bins and optional retention of FASTQs extracted during realignment.
VCF: opt-in Ultima flow-mode GATK scattering. It can change genotypes and annotations and needs 6 GiB RAM per worker; unscattered calling stays the default.
Ancestry and Settings: optional lineage-tree links, persistent logs, an existing temporary-workload volume, atomic main-output publication, and best-effort keep-awake.

AADR 2M is excluded from all: explicitly install its pinned metadata from Library first. It supports GRCh37/38 only; GRCh38 also needs an already-installed forward chain. It does not support CHM13/T2T, and extraction never downloads missing metadata or chains. Use non-identifying AADR sample labels.

CLI setup is native on Windows, without WSL or MSYS2, and GUI settings save real TOML booleans and numbers, including false. Earlier CLI releases also fixed fresh Windows installation, added warnings for invalid configuration values, and improved alignment reproducibility, memory preflight, and native Windows full-genome validation. Tool and platform requirements still vary by workflow.

Download v0.6.6, read the CLI v0.6.6 release notes and ONT limitations, or see the workflow details and prerequisites.

What it forced into the platform

bundle loading from folders and archives;
manifest.json and page schema validation;
setup dry-runs and setup execution;
dynamic page controls and action state;
terminal output panes and process lifecycle handling;
config storage in platform-standard Application Support paths;
localized strings and semantic icon maps;
bundle-specific packaging and app identity.

Those requirements are generic. The WGSExtract app is concrete, but the runtime behavior belongs to GUI for CLI so another CLI tool can reuse the same shell.

Why it shaped the frontend decision

A toy bundle can make any renderer look good. WGSExtract made the comparison harder: the frontend had to carry real commands, real setup, and real release packaging on Windows, macOS, and Linux. One TypeScript app packaged with Tauri was the surface that carried all of it with a single codebase, so it became the only frontend.