{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "8b5ad102",
   "metadata": {
    "language": "markdown"
   },
   "source": [
    "# Raw repertoires to annotation tables\n",
    "\n",
    "Offline, tiny synthetic cohort and wide paired inputs. Repeated source IDs in different samples and a missing beta chain remain visible. Download the complete examples bundle so the shell script and raw inputs are beside this notebook."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ffa396bc",
   "metadata": {
    "language": "python"
   },
   "outputs": [],
   "source": [
    "# Identify the installed CLI and print the fresh kernel environment.\n",
    "import os, sys, subprocess, tempfile, json\n",
    "from pathlib import Path\n",
    "from importlib.metadata import version\n",
    "os.environ[\"POLARS_MAX_THREADS\"] = \"1\"\n",
    "print(\"Python:\", sys.version, \"CLI interpreter:\", sys.executable)\n",
    "for package in (\"vdjmatch\", \"vdjtools\", \"seqtree\", \"polars\"):\n",
    "    print(package, version(package))\n",
    "# Open this notebook from the repository root or the examples directory.\n",
    "examples = next((p for p in (Path.cwd()/\"examples\", Path.cwd())\n",
    "                 if (p/\"inputs/reference.tsv\").is_file()), None)\n",
    "if examples is None:\n",
    "    raise RuntimeError(\"Use the downloaded complete examples bundle, not this notebook alone\")\n",
    "# Deterministic demonstration: no sampling or RNG seeds are used.\n",
    "output = Path(tempfile.mkdtemp(prefix=\"vdjmatch-example-\"))/\"fresh\"\n",
    "output.mkdir()\n",
    "print(\"Fresh output directory:\", output)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2b0e1ddc",
   "metadata": {
    "language": "python"
   },
   "outputs": [],
   "source": [
    "# Run raw inputs with this kernel's installed CLI, on any supported platform.\n",
    "subprocess.run([sys.executable, \"-m\", \"vdjmatch.cli\", \"annotate\",\n",
    "                \"--metadata\", str(examples/\"inputs/samples.tsv\"),\n",
    "                \"--vdjdb\", str(examples/\"inputs/reference.tsv\"), \"--threads\", \"1\",\n",
    "                \"--output-prefix\", str(output/\"cohort\")], check=True, timeout=120)\n",
    "subprocess.run([sys.executable, \"-m\", \"vdjmatch.cli\", \"annotate\",\n",
    "                str(examples/\"inputs/paired.tsv\"), \"--locus\", \"paired\",\n",
    "                \"--vdjdb\", str(examples/\"inputs/reference.tsv\"), \"--threads\", \"1\",\n",
    "                \"--output-prefix\", str(output/\"paired\")], check=True, timeout=120)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1b0c0a3e",
   "metadata": {
    "language": "python"
   },
   "outputs": [],
   "source": [
    "# Analyze only final CLI tables, preserving sample and source identities.\n",
    "import polars as pl\n",
    "for prefix in (\"cohort\", \"paired\"):\n",
    "    queries = pl.read_csv(output/(prefix+\".queries.tsv\"), separator=\"\\t\", infer_schema_length=0)\n",
    "    selections = pl.read_csv(output/(prefix+\".selection.tsv\"), separator=\"\\t\", infer_schema_length=0)\n",
    "    print(prefix, \"query rows:\", queries.height, \"selection rows:\", selections.height)\n",
    "    identity = [c for c in (\"sample_id\", \"source_sequence_id\", \"clone_id\", \"status\", \"has_support\")\n",
    "                if c in selections.columns]\n",
    "    print(selections.select(identity))\n",
    "    manifest = json.loads((output/(prefix+\".manifest.json\")).read_text())\n",
    "    print(\"Native threads:\", manifest[\"native_threads\"])\n",
    "assert pl.read_csv(output/\"cohort.queries.tsv\", separator=\"\\t\").height == 4\n",
    "assert pl.read_csv(output/\"paired.queries.tsv\", separator=\"\\t\").height == 2\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "91c23da3",
   "metadata": {
    "language": "markdown"
   },
   "source": [
    "All displayed results come from fresh CLI tables. Raw synthetic `DEMO_A`/`DEMO_B` labels are software examples, not experimental binding truth. Keep no-hits and missing-chain dispositions. This example establishes neither accuracy nor a false-call threshold. Inspect the CLI manifests beside the tables for model, reference and resource settings. No computed outputs are committed or reused."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "name": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}