sample-notebooks/fhir-vcf-clustering.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "nteract": {
     "transient": {
      "deleting": false
     }
    },
    "tags": []
   },
   "source": [
    "# Clustering analysis use-case with Synthetic FHIR and VCF Data\n",
    "### Overview\n",
    "**Important:** This walk-through assumes you have already used [our previous notebook](https://github.com/microsoft/genomicsnotebook/blob/main/sample-notebooks/fhir_long_read.ipynb)[<sup>[1]</sup>](#r1) to download FHIR data[<sup>[3]</sup>](#r3) in Parquet format[<sup>[4]</sup>](#r4).  \n",
    "A Microsoft Healthcare and Life Sciences blog post accompanies the previous notebook, and can be found [here](https://techcommunity.microsoft.com/t5/healthcare-and-life-sciences/convert-synthetic-fhir-and-pacbio-vcf-data-to-parquet-and/ba-p/3577038)[<sup>[2]</sup>](#r2).\n",
    "\n",
    "In **Section 1**, FHIR data stored in Parquet format is parsed into a pandas DataFrame[<sup>[5]</sup>](#r5) for further analysis.  \n",
    "In **Section 2**, individual patient VCFs are merged into a single joint VCF. We are starting from the raw data again, rather than the Parquet data, in order to combine the VCFs in a different format (a joint VCF).  \n",
    "In **Section 3**, the joint VCF is converted into a pandas DataFrame and missing values are imputed. Since our input is [VCF and not GVCF](https://github.com/broadinstitute/gatk-docs/blob/master/gatk3-faqs/What_is_a_GVCF_and_how_is_it_different_from_a_'regular'_VCF%3F.md)[<sup>[6]</sup>](#r6), absent variant calls are assumed to all be homozygous reference calls of average confidence. In future work, we plan to work with GVCF inputs.  \n",
    "In **Section 4**, the two DataFrames are merged.  \n",
    "In **Section 5**, we perform per-patient clustering analysis.\n",
    "\n",
    "### Contents\n",
    "[1. Extract FHIR Data: from .parquet to DataFrame](#1)  \n",
    "[2. Merge individual VCFs into a joint VCF](#2)  \n",
    "[3. Extract joint VCF: from .vcf to DataFrame](#3)  \n",
    "[4. Data Merging and Normalization](#4)  \n",
    "[5. Clustering Analysis](#5)  \n",
    "[References](#references)\n",
    "\n",
    "Note that we are providing an example architectural design to illustrate how Microsoft tools can be utilized to connect the pieces together (data + interoperability + secure cloud + AI tools), enabling researchers to conduct research analyzing genomics+clinical data. We are not providing or recommending specific instructions for how investigators should conduct their research with this notebook – we will leave that to the professionals!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Extract FHIR Data\n",
    "### from `.parquet` to `DataFrame` <a id=\"1\"></a>\n",
    "First, we'll import the necessary libraries and set some useful variables."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# manage imports\n",
    "import pandas as pd\n",
    "from glob import glob\n",
    "from datetime import datetime\n",
    "import os, json, subprocess\n",
    "\n",
    "# filepaths\n",
    "fhir_dir = \"*/fhir/result\"\n",
    "date = \"2022/07/26\"\n",
    "pb_dir = \"*/pacbio\"\n",
    "\n",
    "# globals\n",
    "max_patients = 50\n",
    "max_variants = 50"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, we'll parse all the FHIR `Patient` data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# parse patient information from each .parquet file\n",
    "fhir_dfs = []\n",
    "for patient_file in glob(f\"{fhir_dir}/Patient/{date}/**/*.parquet\"):\n",
    "    \n",
    "    # create dataframe for this .parquet file\n",
    "    fhir_df = pd.DataFrame()\n",
    "    orig_fhir_df = pd.read_parquet(patient_file)\n",
    "\n",
    "    # set basic fields\n",
    "    fhir_df[\"name\"]    = [f\"{' '.join(name[0]['prefix']) if name[0]['prefix'] else ''} {' '.join(name[0]['given'])} {name[0]['family']}\" for name in orig_fhir_df[\"name\"]]\n",
    "    fhir_df[\"id\"]      = orig_fhir_df[\"id\"]\n",
    "    fhir_df[\"city\"]    = [addr[0][\"city\"]    for addr in orig_fhir_df[\"address\"]]\n",
    "    fhir_df[\"state\"]   = [addr[0][\"state\"]   for addr in orig_fhir_df[\"address\"]]\n",
    "    fhir_df[\"country\"] = [addr[0][\"country\"] for addr in orig_fhir_df[\"address\"]]\n",
    "    fhir_df[\"gender\"]  = orig_fhir_df[\"gender\"]\n",
    "    fhir_df[\"married\"] = [married[\"text\"][0] for married in orig_fhir_df[\"maritalStatus\"]]\n",
    "    fhir_df[\"dead\"]    = [dead is not None for dead in orig_fhir_df[\"deceased\"]]\n",
    "    fhir_df[\"twin+\"]   = [mult[\"boolean\"] if mult[\"boolean\"] is not None else True for mult in orig_fhir_df[\"multipleBirth\"]]\n",
    "    \n",
    "    # set age in years from birthDate\n",
    "    ages = []\n",
    "    for birthdate in orig_fhir_df[\"birthDate\"]:\n",
    "        year, month, day = map(int, birthdate.split(\"-\"))\n",
    "        ages.append((datetime.now()-datetime(year, month, day)).total_seconds() / (60*60*24*365.25))\n",
    "    fhir_df[\"age\"]     = ages\n",
    "\n",
    "    # set \"medical record number\", from a host of other identifiers (social security, etc.)\n",
    "    mrns = []\n",
    "    for identifier_list in orig_fhir_df[\"identifier\"]:\n",
    "        for ident in identifier_list:\n",
    "            if ident['type'] and json.loads(ident['type']['coding'])[0]['code'] == \"MR\":\n",
    "                mrns.append(ident['value'])\n",
    "    fhir_df[\"mrn\"]     = mrns\n",
    "    \n",
    "# merge information\n",
    "    fhir_dfs.append(fhir_df)\n",
    "all_fhir_df = pd.concat(fhir_dfs).reset_index(drop=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For each patient, let's add their newest and oldest medication information based on their FHIR `MedicationRequest` history."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# parse medication information from each .parquet file\n",
    "meds_dfs = []\n",
    "for meds_file in glob(f\"{fhir_dir}/MedicationRequest/{date}/**/*.parquet\"):\n",
    "    \n",
    "    # create a dataframe for this .parquet file\n",
    "    meds_df = pd.DataFrame()\n",
    "    orig_meds_df = pd.read_parquet(meds_file)\n",
    "    \n",
    "    # add simple fields\n",
    "    meds_df[\"id\"] = orig_meds_df[\"id\"]\n",
    "    meds_df[\"code\"] = [\"None\" if med[\"codeableConcept\"] is None else med[\"codeableConcept\"][\"coding\"][0][\"code\"] for med in orig_meds_df[\"medication\"]]\n",
    "    meds_df[\"type\"] = [\"None\" if med[\"codeableConcept\"] is None else med[\"codeableConcept\"][\"text\"] for med in orig_meds_df[\"medication\"]]\n",
    "    meds_df[\"mrn\"] = [subject[\"reference\"].split(':')[2] for subject in orig_meds_df[\"subject\"]] # urn:uuid:<mrn>\n",
    "    \n",
    "    # calculate years elapsed since medication was first requested\n",
    "    elapsed = []\n",
    "    for issued in orig_meds_df[\"authoredOn\"]:\n",
    "        year, month, day = map(int, issued.split(\"T\")[0].split(\"-\"))\n",
    "        elapsed.append((datetime.now()-datetime(year, month, day)).total_seconds() / (60*60*24*365.25))\n",
    "    meds_df[\"elapsed\"] = elapsed\n",
    "\n",
    "# merge information\n",
    "    meds_dfs.append(meds_df)\n",
    "all_meds_df = pd.concat(meds_dfs).reset_index(drop=True)\n",
    "all_meds_df\n",
    "\n",
    "# get newest meds\n",
    "new_meds_df = all_meds_df.loc[all_meds_df.groupby('mrn')['elapsed'].agg('idxmin')]\n",
    "new_codes, new_elapsed, new_types = [], [], []\n",
    "for mrn in all_fhir_df['mrn']:\n",
    "    try:\n",
    "        new_codes.append(new_meds_df[new_meds_df['mrn'] == mrn].iloc[0]['code'])\n",
    "        new_elapsed.append(new_meds_df[new_meds_df['mrn'] == mrn].iloc[0]['elapsed'])\n",
    "        new_types.append(new_meds_df[new_meds_df['mrn'] == mrn].iloc[0]['type'])\n",
    "    except IndexError:\n",
    "        new_codes.append(0)\n",
    "        new_elapsed.append(0)\n",
    "        new_types.append(\"None\")\n",
    "all_fhir_df['newest_med_code'] = new_codes\n",
    "all_fhir_df['newest_med_elapsed'] = new_elapsed\n",
    "all_fhir_df['newest_med_type'] = new_types\n",
    "\n",
    "# get oldest meds\n",
    "old_meds_df = all_meds_df.loc[all_meds_df.groupby('mrn')['elapsed'].agg('idxmax')]\n",
    "old_codes, old_elapsed, old_types = [], [], []\n",
    "for mrn in all_fhir_df['mrn']:\n",
    "    try:\n",
    "        old_codes.append(old_meds_df[old_meds_df['mrn'] == mrn].iloc[0]['code'])\n",
    "        old_elapsed.append(old_meds_df[old_meds_df['mrn'] == mrn].iloc[0]['elapsed'])\n",
    "        old_types.append(old_meds_df[old_meds_df['mrn'] == mrn].iloc[0]['type'])\n",
    "    except IndexError:\n",
    "        old_codes.append(0)\n",
    "        old_elapsed.append(0)\n",
    "        old_types.append(\"None\")\n",
    "all_fhir_df['oldest_med_code'] = old_codes\n",
    "all_fhir_df['oldest_med_elapsed'] = old_elapsed\n",
    "all_fhir_df['oldest_med_type'] = old_types"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We'll also add their newest and oldest diagnoses, from FHIRs `DiagnosticReport` resources."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# parse diagnosis report information from each .parquet file\n",
    "reports_dfs = []\n",
    "for reports_file in glob(f\"{fhir_dir}/DiagnosticReport/{date}/**/*.parquet\"):\n",
    "    \n",
    "    # create a dataframe for this .parquet file\n",
    "    reports_df = pd.DataFrame()\n",
    "    orig_reports_df = pd.read_parquet(reports_file)\n",
    "    \n",
    "    # add simple fields\n",
    "    reports_df[\"id\"]      = orig_reports_df[\"id\"]\n",
    "    reports_df[\"code\"]    = [code[\"coding\"][0][\"code\"] for code in orig_reports_df[\"code\"]]\n",
    "    reports_df[\"type\"]    = [code[\"text\"] for code in orig_reports_df[\"code\"]]\n",
    "    reports_df[\"mrn\"]     = [subject[\"reference\"].split(':')[2] for subject in orig_reports_df[\"subject\"]] # urn:uuid:<mrn>\n",
    "    \n",
    "    # calculate years elapsed since report/diagnosis was issued\n",
    "    elapsed = []\n",
    "    for issued in orig_reports_df[\"issued\"]:\n",
    "        year, month, day = map(int, issued.split(\"T\")[0].split(\"-\"))\n",
    "        elapsed.append((datetime.now()-datetime(year, month, day)).total_seconds() / (60*60*24*365.25))\n",
    "    reports_df[\"elapsed\"] = elapsed\n",
    "    \n",
    "# merge information\n",
    "    reports_dfs.append(reports_df)\n",
    "all_reports_df = pd.concat(reports_dfs).reset_index(drop=True)\n",
    "\n",
    "# get newest reports\n",
    "new_reports_df = all_reports_df.loc[all_reports_df.groupby('mrn')['elapsed'].agg('idxmin')]\n",
    "new_reports_df.head(50)\n",
    "new_codes = []\n",
    "for mrn in all_fhir_df['mrn']:\n",
    "    new_codes.append(new_reports_df[new_reports_df['mrn'] == mrn].iloc[0]['code'])\n",
    "all_fhir_df['newest_report_code'] = new_codes\n",
    "new_elapsed = []\n",
    "for mrn in all_fhir_df['mrn']:\n",
    "    new_elapsed.append(new_reports_df[new_reports_df['mrn'] == mrn].iloc[0]['elapsed'])\n",
    "all_fhir_df['newest_report_elapsed'] = new_elapsed\n",
    "new_types = []\n",
    "for mrn in all_fhir_df['mrn']:\n",
    "    new_types.append(new_reports_df[new_reports_df['mrn'] == mrn].iloc[0]['type'])\n",
    "all_fhir_df['newest_report_type'] = new_types\n",
    "\n",
    "# get oldest reports\n",
    "old_reports_df = all_reports_df.loc[all_reports_df.groupby('mrn')['elapsed'].agg('idxmax')]\n",
    "old_codes = []\n",
    "for mrn in all_fhir_df['mrn']:\n",
    "    old_codes.append(old_reports_df[old_reports_df['mrn'] == mrn].iloc[0]['code'])\n",
    "all_fhir_df['oldest_report_code'] = old_codes\n",
    "old_elapsed = []\n",
    "for mrn in all_fhir_df['mrn']:\n",
    "    old_elapsed.append(old_reports_df[old_reports_df['mrn'] == mrn].iloc[0]['elapsed'])\n",
    "all_fhir_df['oldest_report_elapsed'] = old_elapsed\n",
    "old_types = []\n",
    "for mrn in all_fhir_df['mrn']:\n",
    "    old_types.append(old_reports_df[old_reports_df['mrn'] == mrn].iloc[0]['type'])\n",
    "all_fhir_df['oldest_report_type'] = old_types"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lastly, let's limit the dataset size to the specified maximum number of patients and preview the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>name</th>\n",
       "      <th>id</th>\n",
       "      <th>city</th>\n",
       "      <th>state</th>\n",
       "      <th>country</th>\n",
       "      <th>gender</th>\n",
       "      <th>married</th>\n",
       "      <th>dead</th>\n",
       "      <th>twin+</th>\n",
       "      <th>age</th>\n",
       "      <th>...</th>\n",
       "      <th>newest_med_type</th>\n",
       "      <th>oldest_med_code</th>\n",
       "      <th>oldest_med_elapsed</th>\n",
       "      <th>oldest_med_type</th>\n",
       "      <th>newest_report_code</th>\n",
       "      <th>newest_report_elapsed</th>\n",
       "      <th>newest_report_type</th>\n",
       "      <th>oldest_report_code</th>\n",
       "      <th>oldest_report_elapsed</th>\n",
       "      <th>oldest_report_type</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Mr. Carlos172 Benavides239</td>\n",
       "      <td>f8395477-be71-4542-b241-82d0813c3e1d</td>\n",
       "      <td>Saugus</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>male</td>\n",
       "      <td>M</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>79.612951</td>\n",
       "      <td>...</td>\n",
       "      <td>Captopril 25 MG Oral Tablet</td>\n",
       "      <td>313782</td>\n",
       "      <td>39.845668</td>\n",
       "      <td>Acetaminophen 325 MG Oral Tablet</td>\n",
       "      <td>69409-1</td>\n",
       "      <td>30.742314</td>\n",
       "      <td>U.S. standard certificate of death - 2003 revi...</td>\n",
       "      <td>34117-2</td>\n",
       "      <td>75.169419</td>\n",
       "      <td>None</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Jules135 Schmitt836</td>\n",
       "      <td>eeb892e6-8d4c-4977-a65e-6ec797ee4fa9</td>\n",
       "      <td>Boston</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>male</td>\n",
       "      <td>N</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>0.417879</td>\n",
       "      <td>...</td>\n",
       "      <td>None</td>\n",
       "      <td>0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>None</td>\n",
       "      <td>34117-2</td>\n",
       "      <td>0.149570</td>\n",
       "      <td>None</td>\n",
       "      <td>34117-2</td>\n",
       "      <td>0.417879</td>\n",
       "      <td>None</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Ms. Vonda514 Littel644</td>\n",
       "      <td>475922a6-ab70-45f4-bbaa-35384bf98fe9</td>\n",
       "      <td>Worcester</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>female</td>\n",
       "      <td>N</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>19.079070</td>\n",
       "      <td>...</td>\n",
       "      <td>Ibuprofen 200 MG Oral Tablet</td>\n",
       "      <td>1870230</td>\n",
       "      <td>17.827872</td>\n",
       "      <td>NDA020800 0.3 ML Epinephrine 1 MG/ML Auto-Inje...</td>\n",
       "      <td>34117-2</td>\n",
       "      <td>0.105764</td>\n",
       "      <td>None</td>\n",
       "      <td>34117-2</td>\n",
       "      <td>17.855251</td>\n",
       "      <td>None</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Ms. Marguerita493 Shields502</td>\n",
       "      <td>27127c03-4207-4ab9-9b39-2ee3be2ccb6a</td>\n",
       "      <td>Cambridge</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>female</td>\n",
       "      <td>N</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>23.725202</td>\n",
       "      <td>...</td>\n",
       "      <td>Hydrochlorothiazide 25 MG Oral Tablet</td>\n",
       "      <td>757594</td>\n",
       "      <td>8.740945</td>\n",
       "      <td>Jolivette 28 Day Pack</td>\n",
       "      <td>34117-2</td>\n",
       "      <td>0.056483</td>\n",
       "      <td>None</td>\n",
       "      <td>34117-2</td>\n",
       "      <td>10.654703</td>\n",
       "      <td>None</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Bonita405 Bayer639</td>\n",
       "      <td>a22fb1e0-45e1-4242-b6b5-a792a040e2f5</td>\n",
       "      <td>Chicopee</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>female</td>\n",
       "      <td>N</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>1.633484</td>\n",
       "      <td>...</td>\n",
       "      <td>Amoxicillin 500 MG Oral Tablet</td>\n",
       "      <td>198405</td>\n",
       "      <td>1.222807</td>\n",
       "      <td>Ibuprofen 100 MG Oral Tablet</td>\n",
       "      <td>34117-2</td>\n",
       "      <td>0.196113</td>\n",
       "      <td>None</td>\n",
       "      <td>34117-2</td>\n",
       "      <td>1.633485</td>\n",
       "      <td>None</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 23 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                           name                                    id  \\\n",
       "0    Mr. Carlos172 Benavides239  f8395477-be71-4542-b241-82d0813c3e1d   \n",
       "1           Jules135 Schmitt836  eeb892e6-8d4c-4977-a65e-6ec797ee4fa9   \n",
       "2        Ms. Vonda514 Littel644  475922a6-ab70-45f4-bbaa-35384bf98fe9   \n",
       "3  Ms. Marguerita493 Shields502  27127c03-4207-4ab9-9b39-2ee3be2ccb6a   \n",
       "4            Bonita405 Bayer639  a22fb1e0-45e1-4242-b6b5-a792a040e2f5   \n",
       "\n",
       "        city state country  gender married   dead  twin+        age  ...  \\\n",
       "0     Saugus    MA      US    male       M   True  False  79.612951  ...   \n",
       "1     Boston    MA      US    male       N  False  False   0.417879  ...   \n",
       "2  Worcester    MA      US  female       N  False  False  19.079070  ...   \n",
       "3  Cambridge    MA      US  female       N  False  False  23.725202  ...   \n",
       "4   Chicopee    MA      US  female       N  False  False   1.633484  ...   \n",
       "\n",
       "                         newest_med_type oldest_med_code  oldest_med_elapsed  \\\n",
       "0            Captopril 25 MG Oral Tablet          313782           39.845668   \n",
       "1                                   None               0            0.000000   \n",
       "2           Ibuprofen 200 MG Oral Tablet         1870230           17.827872   \n",
       "3  Hydrochlorothiazide 25 MG Oral Tablet          757594            8.740945   \n",
       "4         Amoxicillin 500 MG Oral Tablet          198405            1.222807   \n",
       "\n",
       "                                     oldest_med_type newest_report_code  \\\n",
       "0                   Acetaminophen 325 MG Oral Tablet            69409-1   \n",
       "1                                               None            34117-2   \n",
       "2  NDA020800 0.3 ML Epinephrine 1 MG/ML Auto-Inje...            34117-2   \n",
       "3                              Jolivette 28 Day Pack            34117-2   \n",
       "4                       Ibuprofen 100 MG Oral Tablet            34117-2   \n",
       "\n",
       "   newest_report_elapsed                                 newest_report_type  \\\n",
       "0              30.742314  U.S. standard certificate of death - 2003 revi...   \n",
       "1               0.149570                                               None   \n",
       "2               0.105764                                               None   \n",
       "3               0.056483                                               None   \n",
       "4               0.196113                                               None   \n",
       "\n",
       "  oldest_report_code  oldest_report_elapsed oldest_report_type  \n",
       "0            34117-2              75.169419               None  \n",
       "1            34117-2               0.417879               None  \n",
       "2            34117-2              17.855251               None  \n",
       "3            34117-2              10.654703               None  \n",
       "4            34117-2               1.633485               None  \n",
       "\n",
       "[5 rows x 23 columns]"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_fhir_df = all_fhir_df.drop(all_fhir_df.index[max_patients:])\n",
    "all_fhir_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Merge individual VCFs into a joint VCF <a id=\"2\"></a>\n",
    "First, we need to install `bcftools` in order to process our VCF files. The `bcftools` documentation[<sup>[7]</sup>](#r7) can be found [here](https://samtools.github.io/bcftools/bcftools.html).  \n",
    "\n",
    "`sudo apt-get install bcftools` is insufficient because the Ubuntu 18.04 package repository contains `bcftools-1.7`, which failed for us on this processing pipeline with an old bug. In the following cell, we install a more recent version, `bcftools-1.15`, from source.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%cd */code\n",
    "!sudo apt-get -y install liblzma-dev libbz2-dev\n",
    "!wget https://github.com/samtools/bcftools/releases/download/1.15.1/bcftools-1.15.1.tar.bz2\n",
    "!tar -xf bcftools-1.15.1.tar.bz2\n",
    "%cd bcftools-1.15.1\n",
    "!sudo make install\n",
    "!bcftools --version"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We must also install `tabix`[<sup>[8]</sup>](#r8), a tool which indexes VCF files for faster processing."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!sudo apt-get install -y tabix"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We're first going to pre-process our VCF files by splitting up sites with multiple alleles into separate records, each with a single allele. For example,\n",
    "\n",
    "CHROM  | POS  | REF  | ALT \n",
    "-------|------|------|----\n",
    "chr20  | 1232 | A    | T,C\n",
    "\n",
    "will become\n",
    "\n",
    "CHROM  | POS  | REF  | ALT  \n",
    "-------|------|------|----\n",
    "chr20  | 1232 | A    | T  \n",
    "chr20  | 1232 | A    | C\n",
    "\n",
    "This ensures that the number of fields is static for each entry, particularly for the `PL` and `AF` columns."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "# skip VCFs which have already been split/pruned\n",
    "orig_vcfs = set(glob(f\"{pb_dir}/**/*.vcf.gz\", recursive=True)) - \\\n",
    "            set(glob(f\"{pb_dir}/**/*_split.vcf.gz\", recursive=True)) - \\\n",
    "            set(glob(f\"{pb_dir}/**/*_pruned.vcf.gz\", recursive=True))\n",
    "\n",
    "for vcf_fn in orig_vcfs:\n",
    "    print(f\"Splitting multi-alleles in '{vcf_fn}'\")\n",
    "    subprocess.run([\n",
    "        \"bcftools\", \"norm\", \n",
    "        \"--multiallelics\", \"-both\",\n",
    "        \"--output-type\", \"z\",\n",
    "        \"--output\", f\"{vcf_fn[:-7]}_split.vcf.gz\", # remove .vcf.gz\n",
    "        vcf_fn\n",
    "    ])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, we perform LD (linkage disequilibrium) pruning[<sup>[9]</sup>](#r9) in order to remove variants with high covariance. Since we're only going to be looking at a small subset of each patient's variants, we want to make sure that the variants we do investigate are fairly independent."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "for vcf_fn in glob(f\"{pb_dir}/**/*_split.vcf.gz\", recursive=True):\n",
    "    print(f\"Pruning variants in '{vcf_fn}'\")\n",
    "    subprocess.run([\n",
    "        \"bcftools\", \"+prune\", \n",
    "        \"-m\", \"0.2\",\n",
    "        vcf_fn,\n",
    "        \"--output-type\", \"z\",\n",
    "        \"--output\", f\"{vcf_fn[:-13]}_pruned.vcf.gz\", # remove _split.vcf.gz\n",
    "    ])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Index all individual VCFs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time \n",
    "for vcf_fn in glob(f\"{pb_dir}/**/*_pruned.vcf.gz\", recursive=True): # no gvcfs\n",
    "    print(f\"Indexing '{vcf_fn}'\")\n",
    "    subprocess.run([\"tabix\", \"-f\", \"-p\", \"vcf\", vcf_fn])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Merge VCF files into single joint VCF, and index it."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "subprocess.run([\"bcftools\", \"merge\"] + \n",
    "        [\"-m\", \"none\"] +\n",
    "        list(glob(f\"{pb_dir}/**/*_pruned.vcf.gz\", recursive=True))[:max_patients] +\n",
    "        [\"-o\", f\"{pb_dir}_joint/{max_patients}_patients.vcf.gz\"]\n",
    ")\n",
    "subprocess.run([\"tabix\", \"-f\", \"-p\", \"vcf\", f\"{pb_dir}_joint/{max_patients}_patients.vcf.gz\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Extract joint VCF <a id=\"3\"></a>\n",
    "### from `.vcf` to `DataFrame`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, we'll use `bcftools query` to convert our VCF to a more general format that works well with existing data science libraries: TSV."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "# convert joint VCF to CSV\n",
    "subprocess.run([\"bcftools\", \"query\", \n",
    "        \"--print-header\",\n",
    "        \"-f\", \"%CHROM\\t%POS\\t%TYPE\\t%REF\\t%ALT\\t%QUAL\\t%FILTER\\t%INFO/DP\\t%INFO/AF\\t%INFO/AQ\\t%INFO/AN\\t%INFO/AC[\\t%GT\\t%AD\\t%DP\\t%GQ\\t%PL]\\n\", \n",
    "        f\"{pb_dir}_joint/{max_patients}_patients.vcf.gz\",\n",
    "        \"-o\", f\"{pb_dir}_tsv/{max_patients}_patients.tsv\"\n",
    "])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, let's define a list of all patients, and functions for computing GT and PL."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save list of patients\n",
    "patients = [p.decode() for p in subprocess.run(\n",
    "    [\"bcftools\", \"query\", \n",
    "         \"--list-samples\", f\"{pb_dir}_joint/{max_patients}_patients.vcf.gz\"], \n",
    "    stdout=subprocess.PIPE).stdout.splitlines()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# replace unknown calls with ref-calls, and remove phasing info (1|0 -> 0/1)\n",
    "def gt_type(gt_str):\n",
    "    gt_str = gt_str.replace(\".\",\"0\")\n",
    "    if (gt_str[0] == '0' and gt_str[2] == '1') or (gt_str[0] == '1' and gt_str[2] == '0'): return '0/1'\n",
    "    if gt_str[0] == '0' and gt_str[2] == '0': return '0/0'\n",
    "    if gt_str[0] == '1' and gt_str[2] == '1': return '1/1'\n",
    "    else: return '?/?'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# parse PL data if available; otherwise compute from other fields\n",
    "def get_pl(gt, gq, pl, ac):\n",
    "    if pl == '.': # missing, compute from GT/GQ\n",
    "        if (ac == 0 and gt == '0/0') or (ac == 1 and gt == '0/1') or (ac == 2 and gt == '1/1'): # chosen allele\n",
    "            return 0\n",
    "        else: # not chosen allele\n",
    "            return gq \n",
    "    else: # parse out of PL data\n",
    "        return int(pl.split(',')[ac])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Afterwards, we load the TSV into a DataFrame, impute missing values, and perform filtering."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "# remove numbered prefix from column names (e.g. \"[1]CHROM\"->\"CHROM\")\n",
    "vcf_df = pd.read_csv(f\"{pb_dir}_tsv/{max_patients}_patients.tsv\", delimiter=\"\\t\", nrows=100000)\n",
    "vcf_df.columns = [col.split(\"]\")[1] for col in vcf_df.columns]\n",
    "\n",
    "# filter out variants which are too homogenous across patients (for more interesting analysis, don't use for real datasets)\n",
    "# a real application may instead have a whitelist/mask of genomic regions or variants of interest\n",
    "min_freq = 0.05\n",
    "max_freq = 0.95\n",
    "gt_is_empty = pd.concat([vcf_df[f'{p}:GT'] == './.' for p in patients], axis=1)\n",
    "gt_is_ref = pd.concat([vcf_df[f'{p}:GT'] == '0/0' for p in patients], axis=1)\n",
    "n_valid = max_patients - (gt_is_empty | gt_is_ref).sum(axis=1)\n",
    "in_freq_range = [x/max_patients < max_freq and x/max_patients > min_freq for x in n_valid]\n",
    "vcf_df = vcf_df[in_freq_range]\n",
    "print(f'selected variants: {len(vcf_df)}')\n",
    "\n",
    "# impute missing values using median for all numeric fields\n",
    "number_cols = [\"QUAL\", \"DP\", \"AQ\"] + [f\"{patient}:GQ\" for patient in patients] + [f\"{patient}:DP\" for patient in patients]\n",
    "print(f'Missing values in the following columns will be imputed using per-patient medians.\\nThis is required because unlike GVCFs, VCFs do not contain depth and quality information for reference calls.')\n",
    "for nc in number_cols:\n",
    "    print(f'{nc} missing: {len(vcf_df[vcf_df[nc] == \".\"])}')\n",
    "    median = pd.to_numeric(vcf_df.loc[vcf_df[nc] != \".\", nc]).median()\n",
    "    vcf_df.loc[vcf_df[nc]=='.', nc] = median\n",
    "    vcf_df[nc] = pd.to_numeric(vcf_df[nc])\n",
    "\n",
    "# compute phred likelihoods and allele frequencies for each sample\n",
    "for p in patients:\n",
    "    vcf_df[f'{p}:GT'] = vcf_df[f'{p}:GT'].map(gt_type)\n",
    "    vcf_df[f'{p}:PL_0/0'] = vcf_df.apply(lambda row: get_pl(gt = row[f'{p}:GT'], gq = row[f'{p}:GQ'], pl = row[f'{p}:PL'], ac = 0), axis=1)\n",
    "    vcf_df[f'{p}:PL_0/1'] = vcf_df.apply(lambda row: get_pl(gt = row[f'{p}:GT'], gq = row[f'{p}:GQ'], pl = row[f'{p}:PL'], ac = 1), axis=1)\n",
    "    vcf_df[f'{p}:PL_1/1'] = vcf_df.apply(lambda row: get_pl(gt = row[f'{p}:GT'], gq = row[f'{p}:GQ'], pl = row[f'{p}:PL'], ac = 2), axis=1)\n",
    "    \n",
    "    vcf_df[f'{p}:AF_0'] = vcf_df.apply(lambda row: 1 if row[f'{p}:AD'] == '.' else int(row[f'{p}:AD'].split(\",\")[0]) / max(row[f'{p}:DP'], 1) , axis=1)\n",
    "    vcf_df[f'{p}:AF_1'] = vcf_df.apply(lambda row: 0 if row[f'{p}:AD'] == '.' else int(row[f'{p}:AD'].split(\",\")[1]) / max(row[f'{p}:DP'], 1) , axis=1)\n",
    "\n",
    "# filter by depth/quality, and remove complex variants\n",
    "vcf_df = vcf_df[vcf_df['DP'] >= 15]\n",
    "vcf_df = vcf_df[vcf_df['QUAL'] >= 20]\n",
    "vcf_df = vcf_df[vcf_df['TYPE'] != 'OTHER']\n",
    "print(f'passing variants: {len(vcf_df)}')\n",
    "\n",
    "# limit to specified number of variants\n",
    "vcf_df = vcf_df.drop(vcf_df.index[max_variants:])\n",
    "vcf_df.reset_index(inplace=True)\n",
    "print(f'selected variants: {len(vcf_df)}')\n",
    "vcf_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lastly, we'll transpose the dataframe so that each row corresponds to a single patient."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "\n",
    "# define the fields we'll be using in our final dataframe\n",
    "fields = []\n",
    "sample_fields = [\"GT\", \"AF_0\", \"AF_1\", \"PL_0/0\", \"PL_0/1\", \"PL_1/1\", \"DP\"]\n",
    "\n",
    "# create empty dataframe\n",
    "pb_df = pd.DataFrame(columns=[f\"{var_id}:{fld}\" for var_id in range(len(vcf_df)) for fld in fields + sample_fields], index=patients)\n",
    "\n",
    "# fill dataframe\n",
    "for idx, row in vcf_df.iterrows():\n",
    "    for f in fields:\n",
    "        for p in patients:\n",
    "            pb_df.loc[p][f\"{idx}:{f}\"] = row[f]\n",
    "    for f in sample_fields:\n",
    "        for p in patients:\n",
    "            pb_df.loc[p][f\"{idx}:{f}\"] = row[f\"{p}:{f}\"]\n",
    "\n",
    "# new index\n",
    "pb_df.reset_index(inplace=True)\n",
    "pb_df = pb_df.rename(columns = {'index': 'patient_id'})\n",
    "pb_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Data Merging and Normalization <a id=\"4\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Since one row corresponds to a single patient in both our PacBio and FHIR datasets, we can easily append these two datasets side by side.  \n",
    "Note that for real applications, each row's PacBio data must be matched up with the FHIR data for the corresponding patient. Since our FHIR data is synthetic, this matchup doesn't matter."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>name</th>\n",
       "      <th>id</th>\n",
       "      <th>city</th>\n",
       "      <th>state</th>\n",
       "      <th>country</th>\n",
       "      <th>gender</th>\n",
       "      <th>married</th>\n",
       "      <th>dead</th>\n",
       "      <th>twin+</th>\n",
       "      <th>age</th>\n",
       "      <th>...</th>\n",
       "      <th>48:PL_0/1</th>\n",
       "      <th>48:PL_1/1</th>\n",
       "      <th>48:DP</th>\n",
       "      <th>49:GT</th>\n",
       "      <th>49:AF_0</th>\n",
       "      <th>49:AF_1</th>\n",
       "      <th>49:PL_0/0</th>\n",
       "      <th>49:PL_0/1</th>\n",
       "      <th>49:PL_1/1</th>\n",
       "      <th>49:DP</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Mr. Carlos172 Benavides239</td>\n",
       "      <td>f8395477-be71-4542-b241-82d0813c3e1d</td>\n",
       "      <td>Saugus</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>male</td>\n",
       "      <td>M</td>\n",
       "      <td>True</td>\n",
       "      <td>False</td>\n",
       "      <td>79.612951</td>\n",
       "      <td>...</td>\n",
       "      <td>8</td>\n",
       "      <td>8</td>\n",
       "      <td>7</td>\n",
       "      <td>0/0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>8</td>\n",
       "      <td>8</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Jules135 Schmitt836</td>\n",
       "      <td>eeb892e6-8d4c-4977-a65e-6ec797ee4fa9</td>\n",
       "      <td>Boston</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>male</td>\n",
       "      <td>N</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>0.417879</td>\n",
       "      <td>...</td>\n",
       "      <td>3</td>\n",
       "      <td>3</td>\n",
       "      <td>7</td>\n",
       "      <td>0/0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>3</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Ms. Vonda514 Littel644</td>\n",
       "      <td>475922a6-ab70-45f4-bbaa-35384bf98fe9</td>\n",
       "      <td>Worcester</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>female</td>\n",
       "      <td>N</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>19.079070</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0/1</td>\n",
       "      <td>0.285714</td>\n",
       "      <td>0.714286</td>\n",
       "      <td>29</td>\n",
       "      <td>0</td>\n",
       "      <td>23</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Ms. Marguerita493 Shields502</td>\n",
       "      <td>27127c03-4207-4ab9-9b39-2ee3be2ccb6a</td>\n",
       "      <td>Cambridge</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>female</td>\n",
       "      <td>N</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>23.725202</td>\n",
       "      <td>...</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>5</td>\n",
       "      <td>0/0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Bonita405 Bayer639</td>\n",
       "      <td>a22fb1e0-45e1-4242-b6b5-a792a040e2f5</td>\n",
       "      <td>Chicopee</td>\n",
       "      <td>MA</td>\n",
       "      <td>US</td>\n",
       "      <td>female</td>\n",
       "      <td>N</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td>1.633484</td>\n",
       "      <td>...</td>\n",
       "      <td>9</td>\n",
       "      <td>9</td>\n",
       "      <td>15</td>\n",
       "      <td>0/0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>9</td>\n",
       "      <td>9</td>\n",
       "      <td>15</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 374 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                           name                                    id  \\\n",
       "0    Mr. Carlos172 Benavides239  f8395477-be71-4542-b241-82d0813c3e1d   \n",
       "1           Jules135 Schmitt836  eeb892e6-8d4c-4977-a65e-6ec797ee4fa9   \n",
       "2        Ms. Vonda514 Littel644  475922a6-ab70-45f4-bbaa-35384bf98fe9   \n",
       "3  Ms. Marguerita493 Shields502  27127c03-4207-4ab9-9b39-2ee3be2ccb6a   \n",
       "4            Bonita405 Bayer639  a22fb1e0-45e1-4242-b6b5-a792a040e2f5   \n",
       "\n",
       "        city state country  gender married   dead  twin+        age  ...  \\\n",
       "0     Saugus    MA      US    male       M   True  False  79.612951  ...   \n",
       "1     Boston    MA      US    male       N  False  False   0.417879  ...   \n",
       "2  Worcester    MA      US  female       N  False  False  19.079070  ...   \n",
       "3  Cambridge    MA      US  female       N  False  False  23.725202  ...   \n",
       "4   Chicopee    MA      US  female       N  False  False   1.633484  ...   \n",
       "\n",
       "  48:PL_0/1 48:PL_1/1  48:DP 49:GT   49:AF_0   49:AF_1 49:PL_0/0 49:PL_0/1  \\\n",
       "0         8         8      7   0/0         1         0         0         8   \n",
       "1         3         3      7   0/0         1         0         0         3   \n",
       "2         0         0      5   0/1  0.285714  0.714286        29         0   \n",
       "3         4         4      5   0/0         1         0         0         4   \n",
       "4         9         9     15   0/0         1         0         0         9   \n",
       "\n",
       "   49:PL_1/1 49:DP  \n",
       "0          8     7  \n",
       "1          3     7  \n",
       "2         23     7  \n",
       "3          4     5  \n",
       "4          9    15  \n",
       "\n",
       "[5 rows x 374 columns]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_df = pd.concat([all_fhir_df, pb_df], axis=1)\n",
    "all_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Once all this information is in a single table, we can select the most useful columns and convert/normalize data into a format which is best for ML applications. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>alive</th>\n",
       "      <th>twin+</th>\n",
       "      <th>male</th>\n",
       "      <th>age</th>\n",
       "      <th>married</th>\n",
       "      <th>never_married</th>\n",
       "      <th>single</th>\n",
       "      <th>old_report_age</th>\n",
       "      <th>new_report_age</th>\n",
       "      <th>old_med_age</th>\n",
       "      <th>...</th>\n",
       "      <th>46:0/1</th>\n",
       "      <th>46:1/1</th>\n",
       "      <th>47:0/0</th>\n",
       "      <th>47:0/1</th>\n",
       "      <th>47:1/1</th>\n",
       "      <th>48:0/0</th>\n",
       "      <th>48:0/1</th>\n",
       "      <th>48:1/1</th>\n",
       "      <th>49:0/0</th>\n",
       "      <th>49:0/1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.944276</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.956858</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.580042</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.003922</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.222505</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0.223207</td>\n",
       "      <td>0.002496</td>\n",
       "      <td>0.259524</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.277903</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0.131037</td>\n",
       "      <td>0.000891</td>\n",
       "      <td>0.127244</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.014494</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0.015560</td>\n",
       "      <td>0.005438</td>\n",
       "      <td>0.017801</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 148 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   alive  twin+  male       age  married  never_married  single  \\\n",
       "0      0      0     1  0.944276        1              0       0   \n",
       "1      1      0     1  0.000000        0              1       0   \n",
       "2      1      0     0  0.222505        0              1       0   \n",
       "3      1      0     0  0.277903        0              1       0   \n",
       "4      1      0     0  0.014494        0              1       0   \n",
       "\n",
       "   old_report_age  new_report_age  old_med_age  ...  46:0/1  46:1/1  47:0/0  \\\n",
       "0        0.956858        1.000000     0.580042  ...       0       0       1   \n",
       "1        0.000000        0.003922     0.000000  ...       0       0       1   \n",
       "2        0.223207        0.002496     0.259524  ...       0       0       1   \n",
       "3        0.131037        0.000891     0.127244  ...       0       0       1   \n",
       "4        0.015560        0.005438     0.017801  ...       0       0       1   \n",
       "\n",
       "   47:0/1  47:1/1  48:0/0  48:0/1  48:1/1  49:0/0  49:0/1  \n",
       "0       0       0       1       0       0       1       0  \n",
       "1       0       0       1       0       0       1       0  \n",
       "2       0       0       0       1       0       0       1  \n",
       "3       0       0       1       0       0       1       0  \n",
       "4       0       0       1       0       0       1       0  \n",
       "\n",
       "[5 rows x 148 columns]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# for normalization\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "mms = MinMaxScaler()\n",
    "\n",
    "# select useful FHIR columns\n",
    "ml_df = pd.DataFrame()\n",
    "ml_df['alive'] = [1-int(x) for x in all_df['dead']]\n",
    "ml_df['twin+'] = [int(x) for x in all_df['twin+']]\n",
    "ml_df['male'] = [1 if g == 'male' else 0 for g in all_df['gender']]\n",
    "ml_df['age'] = all_df['age']\n",
    "ml_df = pd.merge(\n",
    "    left = ml_df,\n",
    "    right = pd.get_dummies(all_df['married']),\n",
    "    left_index = True,\n",
    "    right_index = True,\n",
    ")\n",
    "ml_df = ml_df.rename(columns={'M': 'married', 'N': 'never_married', 'S': 'single'})\n",
    "ml_df['old_report_age'] = all_df['oldest_report_elapsed']\n",
    "ml_df['new_report_age'] = all_df['newest_report_elapsed']\n",
    "ml_df['old_med_age'] = all_df['oldest_med_elapsed']\n",
    "ml_df['new_med_age'] = all_df['newest_med_elapsed']\n",
    "new_med_code_df = pd.get_dummies(all_df['newest_med_code'])\n",
    "ml_df = pd.merge(\n",
    "    left = ml_df,\n",
    "    right = new_med_code_df.add_prefix(\"new_med_code:\"),\n",
    "    left_index = True,\n",
    "    right_index = True,\n",
    ")\n",
    "\n",
    "# add all variants (ignoring DP, AF, PL for now)\n",
    "for var_idx in range(max_variants):\n",
    "    gt_df = pd.get_dummies(all_df[f'{var_idx}:GT'])\n",
    "    ml_df = pd.merge(\n",
    "        left = ml_df,\n",
    "        right = gt_df.add_prefix(f'{var_idx}:'),\n",
    "        left_index = True,\n",
    "        right_index = True,\n",
    "    )\n",
    "\n",
    "    #ml_df[f'{var_idx}:DP'] = all_df[f'{var_idx}:DP']\n",
    "    \n",
    "# normalize\n",
    "age_cols = ['age', 'old_report_age', 'new_report_age', 'old_med_age', 'new_med_age'] #+ [f'{var_idx}:DP' for var_idx in range(max_variants)]\n",
    "ml_df[age_cols] = mms.fit_transform(ml_df[age_cols])\n",
    "ml_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 5. Clustering Analysis <a id=\"5\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Before we can perform clustering, we must select a reasonable number of clusters for grouping patients. The k-means++ and spectral clustering algorithms require a predetermined number of clusters as inputs. This can be done using the \"elbow method\", where we select the number of clusters after which there are diminishing returns (in terms of cluster inertia) of introducing additional clusters. This appears in a graph of (x = # clusters, y = total inertia) as a bend, or elbow. In our dataset, we find a slight elbow at `n_clusters = 4`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "from sklearn.cluster import KMeans\n",
    "\n",
    "inertias = []\n",
    "for i in range(1,11):\n",
    "    kmeans = KMeans(n_clusters = i).fit(ml_df)\n",
    "    inertias.append(kmeans.inertia_)\n",
    "\n",
    "plt.plot(range(1,11), inertias, marker='o')\n",
    "plt.title('\"Elbow\" Method')\n",
    "plt.xlabel('Number of Clusters')\n",
    "plt.ylabel('Inertia')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here, we demonstrate several clustering methods on our patient dataset.  \n",
    "You can find more information about each clustering method at [`scikit-learn`'s Clustering Page](https://scikit-learn.org/stable/modules/clustering.html#clustering)[<sup>[10]</sup>](#r10)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.decomposition import PCA\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.cluster import KMeans, DBSCAN, SpectralClustering\n",
    "\n",
    "# three clustering options\n",
    "n_clusters = 4\n",
    "cluster1 = KMeans(n_clusters=n_clusters, init='k-means++', n_init=5).fit(ml_df)\n",
    "cluster2 = DBSCAN(eps=3, min_samples=3).fit(ml_df)\n",
    "cluster3 = SpectralClustering(n_clusters=n_clusters).fit(ml_df)\n",
    "\n",
    "# break down ML DF into top 2 principal components\n",
    "pca2 = PCA(n_components = 2)\n",
    "pc = pca2.fit_transform(ml_df)\n",
    "pc_df = pd.DataFrame(data = pc, columns = ['pc1', 'pc2'])\n",
    "\n",
    "# plot different clustering methods\n",
    "fig, ax = plt.subplots(1,3, figsize=(15,5))\n",
    "methods = [\"K-Means++\", \"DBSCAN\", \"Spectral\"]\n",
    "clusters = [cluster1, cluster2, cluster3]\n",
    "for x in range(3):\n",
    "    ax[x].set_title(f\"{methods[x]} Clustering\")\n",
    "    ax[x].set_xlabel('Principal Component #1')\n",
    "    ax[x].set_ylabel('Principal Component #2')\n",
    "    ax[x].scatter(pc_df['pc1'], pc_df['pc2'], c=clusters[x].labels_)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, let's evaluate the performance of these clustering methods. More information the metrics used can again be found at [`scikit-learn`'s Clustering Page](https://scikit-learn.org/stable/modules/clustering.html#clustering-performance-evaluation)[<sup>[10]</sup>](#r10)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import davies_bouldin_score, calinski_harabasz_score, silhouette_score\n",
    "from  matplotlib.colors import LinearSegmentedColormap\n",
    "import numpy as np\n",
    "\n",
    "# calculate evaluation metrics\n",
    "metrics = [\"Davies-Bouldin\", \"Calinski-Harabasz\", \"Silhouette\"]\n",
    "high_is_good = [False, True, True]\n",
    "scores = np.zeros((3,3))\n",
    "for x in range(3):\n",
    "    scores[x,0] = davies_bouldin_score(ml_df, clusters[x].labels_)\n",
    "    scores[x,1] = calinski_harabasz_score(ml_df, clusters[x].labels_)\n",
    "    scores[x,2] = silhouette_score(ml_df, clusters[x].labels_, metric='euclidean')\n",
    "\n",
    "# plot colormaps\n",
    "fig, ax = plt.subplots(1,3, figsize=(6,5))\n",
    "for x in range(3):\n",
    "    cmap = LinearSegmentedColormap.from_list('rg',[\"r\", \"y\", \"g\"], N=256) if high_is_good[x] else \\\n",
    "           LinearSegmentedColormap.from_list('gr',[\"g\", \"y\", \"r\"], N=256)\n",
    "    ax[x].matshow(np.expand_dims(scores[:,x], axis=1), cmap=cmap)\n",
    "    ax[x].set_xticks([0])\n",
    "    ax[x].set_xticklabels([metrics[x]])\n",
    "    if x == 0:\n",
    "        ax[x].set_yticklabels([0] + methods)\n",
    "    else:\n",
    "        ax[x].set_yticks([])\n",
    "    for y in range(3):\n",
    "        ax[x].text(0, y, f'{scores[y,x]:2f}', va='center', ha='center')\n",
    "\n",
    "# display\n",
    "ax[0].set_ylabel('Clustering Methods')\n",
    "ax[1].set_xlabel('Metrics')\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lastly, here's an agglomerative clustering dendrogram, which can also be used to cluster patients:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABG0AAAE9CAYAAACx2J+XAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nO3df5hld10n+PeXdFDYqsho9Y+a7gc7cTeQCmlDyCQiRgNmxqggGxWzRB0n88OwM87sOHFXdtZdxN19Jo+7cdbxGSWsMxHQMBmQGAIYIJDG0GLaJLbdpBvaNd0w3XZ3ukDsLoMQwnf/uLdCpelO36pzTt1zu1+v5+mnbt2c+pxvzj33e8593+/5nlJrDQAAAAD98pxxNwAAAACArye0AQAAAOghoQ0AAABADwltAAAAAHpIaAMAAADQQ0IbAAAAgB5as5yFZ2Zm6ubNmztqCgAAAMDZ5+GHH56vta498fllhTabN2/OQw891F6rAAAAAM5ypZTPnOx5l0cBAAAA9JDQBgAAAKCHhDYAAAAAPSS0AQAAAOghoQ0AAABADwltAAAAAHpIaAMAAADQQ0IbAAAAgB4S2gAAAAD0kNAGAAAAoIfWjLsBo7rjwc/m7h0Hx90MAE7w2ks35oYrXzjuZgAAwBlnYkba3L3jYHYfOjbuZgCwxO5DxwTqAADQkYkZaZMkc7Pn5c6bXj7uZgAwdP1tnxh3EwAA4Iw1MSNtAAAAAM4mQhsAAACAHpqoy6M4e5h4GibD4lxjLpOC/jNpOABMHiNt6CUTT8NkmJs9L3Oz5427GcBpmDQcACaTkTb0lomnAaAdRsMBwGQy0gYAAACgh4Q2AAAAAD0ktAEAAADoIaENAAAAQA8JbQAAAAB6yN2jAOiFOx78rFsSQ0d2HzqWxF2koCuvvXRjbrjyheNuBnAGMtIGgF64e8fBpz9YAu2amz0vc7PnjbsZcEbafeiYLx2AzhhpA0BvzM2elztvevm4mwEAIzOCDeiSkTYAAAAAPSS0AQAAAOghoQ0AAABADwltAAAAAHpIaAMAAADQQ0IbAAAAgB4S2gAAAAD0kNAGAAAAoIeENgAAAAA9JLQBAAAA6CGhDQAAAEAPCW0AAAAAekhoAwAAANBDQhsAAACAHhLaAAAAAPSQ0AYAAACgh4Q2AAAAAD0ktAEAAADoIaENAAAAQA8JbQAAAAB6SGgDAAAA0ENCGwAAAIAeEtoAAAAA9JDQBgAAAKCHhDYAAAAAPSS0AQAAAOghoQ0AAABADwltAAAAAHpIaAMAAADQQ0IbAAAAgB4S2gAAAAD0kNAGAAAAoIeENgAAAAA9JLQBAAAA6KE1424AAAAwHnc8+NncvePguJsx0XYfOpYkuf62T4y5JZPttZduzA1XvnDczYDeMdIGAADOUnfvOPh06MDKzM2el7nZ88bdjIm2+9Ax4SGcgpE2AABwFpubPS933vTycTeDs5hRSnBqRtoAAAAA9JDQBgAAAKCHXB4FnJZJClkNJnJktZjsEgCYFEIbGunqw3yXH96crC/f4iSFJtmjS/YvVsPi8cVxAACYBEIbGunqw3xXH96crK+cSQqBM4GRXADAJBHa0NgkfZh3sg4AAMCkMBExAAAAQA8JbQAAAAB6SGgDAAAA0ENCGwAAAIAeEtoAAAAA9JDQBgAAAKCHhDYAAAAAPSS0AQAAAOghoQ0AAABADwltAAAAAHpIaAMAAADQQ2vG3QAAAIAzzR0PfjZ37zg47mZMhN2HjiVJrr/tE2NuyWR47aUbc8OVLxx3M1glRtoAAAC07O4dB58OI3h2c7PnZW72vHE3YyLsPnRMGHiWMdIGAACgA3Oz5+XOm14+7mZwBjEa6exjpA0AAABADwltAAAAAHpIaAMAAADQQ0IbAAAAgB4S2gAAAAD0kNAGAAAAoIeENgAAAAA9JLQBAAAA6KE1424AAADw9RYePJQndhztdB1PHvrrJMnjt+3sdD3Pv3Rtpq6c7XQdAGcioQ0AAPTQEzuO5slDCzl3dqqzdbxldl1ntRc9eWghTyRCG4AVENoAAEBPnTs7lXU3bRl3MxrpehQPwJlMaAMAtGbnffdmz7at427GKR390iVJkjvffPeYW3JqF73i6my55tpxNwMA6AGhDQDQmj3btubo/n1Zu/n8cTflpG76hl3jbsKzOrp/X5IIbQCAJF2FNg/dnux6d7s1D7928PP2/6Pdupf8aHL5je3WBICz2NrN5+f6N90y7mZMpDvf/MZxNwEA6JFuQptd704O70o2XNJayTtf2MEw5sPDb9uENgAAAEDPdHd51IZLkhvf31n5Vtz+g+NuAQAAAMBJPWfcDQAAAADg6wltAAAAAHpIaAMAAADQQ5N1y++270p1eOfgZ1tz27gTFQAAwGkdPPjOHD5yz7ibMXEWFq5Jkjz8yK+NuSWTacP612TjxtePuxnLMlmhTdt3pdqwpZ06iTtRAUCP7Lzv3uzZtnXczVi2x/c/lmRyb/190SuuzpZrrh13M4AJcPjIPVlY2J2pqblxN2Wi/NIr7xt3EybWwsLuHE6ENp3r612p3IkKAHpjz7atObp/X9ZuPn/cTVmWdZsvGHcTVuzo/n1JIrQBRjY1NZeXXXbHuJvBWeLhR24YdxNWZPJCGwCAEazdfH6uf9Mt427GWWNSRwcBQJ8JbXiGd+19Vz7w2AdGXv7Tn/+eJMmN9751pOV/4IIfyOsufN2K2nYqf3nnf86x971vpGW/NPPKJMlnfvI3Rlr+vFe/On/r+h9bcdsAAABgpYQ2PMMHHvtAPv35T+dF3/yikZZ/6Us/NnLtT3/+00nSemhz7H3vy9986lP5xhe/+LTL/ur8/SPX/ZtPfSpJhDYAAACMhdCGr/Oib35Rbr/29tbr3nhvd5M0f+OLX5xvfcfbW635mZ/8+63WAwAAgOV4zrgbAAAAAMDXE9oAAAAA9JDQBgAAAKCHhDYAAAAAPSS0AQAAAOghoQ0AAABADwltAAAAAHpIaAMAAADQQ0IbAAAAgB4S2gAAAAD00JpxNwB4do8+cDB7tx8Zaxvmv7CQJLnr1kfG2o4Lr1ifi6/aONY2AAAArBahDfTc3u1HMn9gITObpsbWhp95wbeMbd2L5g8MgiOhDQAAnF0OHnxnDh+5p1GN48d3J0kefuSGFdfYsP412bjx9Y3asVxCG5gAM5umct3Nl427GWM17lE+AADAeBw+ck8WFnZnampuxTWmp1f+t0mysLA7hxOhDQAAAMBSU1Nzedlld4xt/U1G6DQhtAEAAKCVS1BG1calKss1jktboCl3jwIAAODpS1BWw/T0XOPLVZZjYWH3qgVS0CYjbQAAAEgy/ktQujKuS1ugKaENAABwSgsPHsoTO46u+O+fPDS4A+Tjt+1s1I7nX7o2U1fONqoBMGlcHgUAAJzSEzuOPh28rMS5s1M5d3aqURuePLTQKDgCmFRG2gAAAM/q3NmprLtpy9jW33SUDsCkEtoAp/XoAwezd/uRsbZh/sDxJMldtz4y1nZceMX6XHzVxrG2AQAAODsIbYDT2rv9SOYPLGRmU7OhzU3MbJoe27oXzR8YDA0X2gDjsPO+e7Nn29ZxN+OUHt//WJLkzje/ccwteXYXveLqbLnm2nE3AwBGIrQBRjKzaSrX3XzZuJsxVuMe5QOc3fZs25qj+/dl7ebzx92Uk1q3+YJxN+G0ju7flyRCGwAmhtAGAGBCrN18fq5/0y3jbsbE6vsoIAA4kbtHAQAAAPSQ0AYAAACgh4Q2AAAAAD0ktAEAAADoIRMRAxPj0QcOZu/2I2Nb//yB40nGexepC69Y75bjAABwljDSBpgYe7cfyfyBhbGtf2bTdGY2TY9t/fMHFsYaWgEAAKvLSBtgosxsmsp1N1827maMxThH+AAAAKtPaAMAAKy6hQcP5YkdR0da9slDg5G2j9+2c6Tln3/p2kxdObvitgH0hcujAACAVffEjqNPhzGnc+7sVM6dnRpp2ScPLYwcBgH0nZE2AADAWJw7O5V1N21pteaoo3EAJoHQBgCYCDvvuzd7tm0dadnH9z+WJLnzzW8cafmLXnF1tlxz7UqbBgDQCaENjMFybl293NtMuyU0cKbas21rju7fl7Wbzz/tsus2XzBy3aP79yWJ0AY4rYceeii7du0aadnDh1+QJLn99ttHWv6SSy7J5ZdfvuK2AWcmoQ2MweKtq2c2nf7a7OXcYnrxdthCG+BMtXbz+bn+Tbe0WnPU0TgAu3btyuHDh7Nhw4bTLvsPXviFkesePnw4SYQ2wNcR2ix66PZk17tX/veHh9fO3v6DzdpxyY8ml9/YrAYToYtbV7slNABAtzZs2JAbb2z3fH3U0TjA2cfdoxbtendyeLShjie1YcvgXxOHdzULjgAAAIAzhpE2S224JLnx/eNbf9NROgAAABPs4MF35vCRe1qve/z47iTJw4/c0GrdDetfk40bX99qTVjKSBsAAAB64fCRe7KwsLv1utPTc5menmu15sLC7k4CJljKSBsAAAB6Y2pqLi+77I5xN+O02h61AydjpA0AAABADxlpA7AMjz5wMHu3HxnLuucPHE8ynruEXXjFereSBwCAVWakDcAy7N1+JPMHFsay7plN05nZNL3q650/sDC2oAoAAM5mRtp07aHbR7+N9+Gdg5+j3kXqkh9NLr9xZe0CVmxm01Suu/mycTdj1YxjZA/AKHbed2/2bNs68vKP738sSXLnm9840vIXveLqbLnm2pU07ZQWHjyUJ3YcHWnZJw8NviR4/LadIy3//EvXZurK2RW3DYD+MdKma7venRzeNdqyG7YM/o3i8K7RwyAAgDPQnm1bc3T/vpGXX7f5gqzbfMFIyx7dv29ZgdConthx9Okw5nTOnZ3KubNTIy375KGFkcMgACaHkTarYcMlyY3vb7fmqKNxAADOYGs3n5/r33RL63VHHY2zEufOTmXdTSN+UTeiUUfjADBZjLQBAAAA6CEjbQAAOrDc+VZOZ7nzsZxOF/O1AADtMtIGAKADy51v5XSWMx/L6XQ1XwsA0C4jbQAAOtLVfCtNdTlfCwDQHqENwAR69IGD2bv9yKqsa/7A8SSre+vvC69Yn4uv2rhq6wMAgD5yeRTABNq7/UjmD4x2y9imZjZNZ2bT9KqsK0nmDyysWiAFAAB9ZqQNwISa2TSV626+bNzNaN1qjugBgDPZwYPvzOEj94y8/PHju5MkDz9yw0jLb1j/mmzc+PoVtQ0YjZE2AAAAZ6DDR+7JwsLukZefnp7L9PTcSMsuLOxeViAErIyRNgAAAGeoqam5vOyyO1qvO+poHKAZI20AAAAAeshIGwAAAM5oy53fZxTLnQNoVOYKYikjbQAAADijLXd+n1EsZw6gUZkriBMZaQMAAMAZr6v5fdpkriBOZKQNAAAAQA8ZaQMAMGF23ndv9mzbuuK/f3z/Y0mSO9/8xkbtuOgVV2fLNdc2qgEAnJrQBgBgwuzZtjVH9+/L2s3nr+jv122+oHEbju7flyRCGwAm1nImqF7JxNNtTCottAEAmEBrN5+f6990y9jW33SUDgCM2+IE1VNTp59QermTTi8s7M7hRGgDTLZHHziYvduPjLTs/IHjSZK7bn1kpOUvvGJ9Lr5q44rbBgAAnNm6mqC6rUmlTUQMjNXe7Ucyf2BhpGVnNk1nZtP0SMvOH1gYOQwCAADoIyNtgLGb2TSV626+rNWao47GAQAA6CuhDQArspxL25ZjuZfBLYdL5gCAti1nMtvTWclkt6fTxmS4jI/LowBYkeVc2rYcy7kMbjlcMgcAdGFxMts2TE/PLXvC22ezsLC7tUCJ8TDSBoAV6+LStq64ZA4A6EpXk9k21eaIHcbDSBsAAACAHhLaAAAAAPSQ0AYAAACgh8xpAwBAkmTnffdmz7atIy37+P7HkiR3vvmNIy1/0SuuzpZrrl1p0wDgrGSkDQAASZI927bm6P59Iy27bvMFWbf5gpGWPbp/38hhEADwNUbaAADwtLWbz8/1b7ql1ZqjjsYB4GsOHnxn49t1Hz8+uBV5k7tIbVj/mmzc+PpG7WDljLQBAACAnjl85J4sLOxuVGN6ei7T03Mr/vuFhd2NgyOaMdIGAAAAemhqai4vu+yOVmsuZwRPrYPROssZqWNkTruMtAEAAICzxHJG8Cx3pI6ROe0z0gaA3nn0gYPZu/1IqzXnDxxPktx16yOt1bzwivW5+KqNrdUDAFgNXYzgSZrNncPJCW0A6J29249k/sBCZjZNtVZzZtN0/vqvvvR0eNPUl7/4VOYPHG81XBICAQCwlNAGgF6a2TSV626+rNWad936SL54/MlWw6C2zB9YSBKhDQAATxPaAHBW6SIMakObl20BAHBmMBExAAAAQA8JbQAAAAB6SGgDAAAA0ENCGwAAAIAeEtoAAAAA9JDQBgAAAKCHhDYAAAAAPSS0AQAAAOghoQ0AAABAD60ZdwMAGHj0gYPZu/3ISMvOHzieJLnr1kdGWv7CK9bn4qs2rrhtAADA6jPSBqAn9m4/kvkDCyMtO7NpOjObpkdadv7AwshhEAAA0B9G2gD0yMymqVx382Wt1hx1NA4AANAvRtoAAAAA9JDQBgAAAKCHXB4FAA0tZxLpU1nu5NKnYtJpAIAzh5E2ANDQciaRPpXlTC59KiadBgA4sxhpAwAt6GIS6eUy6TQAwJnFSBsAAACAHhLaAAAAAPSQ0AYAAACgh4Q2AAAAAD0ktAEAAADoIaENAAAAQA8JbQAAAAB6SGgDAAAA0ENCGwAAAIAeEtoAAAAA9JDQBgAAAKCHhDYAAAAAPbRm3A0AAE7t0QcOZu/2IyMtO3/geJLkrlsfGWn5C69Yn4uv2rjitgEA0C0jbQCgx/ZuP5L5AwsjLTuzaTozm6ZHWnb+wMLIYRAAAONhpA0A9NzMpqlcd/NlrdYcdTQOAADjY6QNAAAAQA8JbQAAAAB6SGgDAAAA0ENCGwAAAIAeEtoAAAAA9JDQBgAAAKCHhDYAAAAAPSS0AQAAAOghoQ0AAABADwltAAAAAHpIaAMAAADQQ0IbAAAAgB4S2gAAAAD0kNAGAAAAoIeENgAAAAA9JLQBAAAA6CGhDQAAAEAPCW0AAAAAekhoAwAAANBDQhsAAACAHhLaAAAAAPSQ0AYAAACgh4Q2AAAAAD0ktAEAAADoIaENAAAAQA8JbQAAAAB6SGgDAAAA0ENCGwAAAIAeEtoAAAAA9JDQBgAAAKCHhDYAAAAAPSS0AQAAAOghoQ0AAABADwltAAAAAHpIaAMAAADQQ0IbAAAAgB4S2gAAAAD0kNAGAAAAoIeENgAAAAA9JLQBAAAA6CGhDQAAAEAPCW0AAAAAekhoAwAAANBDQhsAAACAHhLaAAAAAPRQqbWOvnApR5N8prvmAAAAAJx1vrXWuvbEJ5cV2gAAAACwOlweBQAAANBDQhsAAACAHhLaAAAAAPSQ0AYAAACgh4Q2AAAAAD0ktAEAAADooTXjbsCZqJTysiQvT/KCJF9I8ke11oca1vyhJPfVWp9ooYkTr5RySZLvzGAbH0nywVrrofG2iiZKKX+n1vrH424H4zPsO/9Lks8leXWSL9ZaPzTeVo2ulPLPaq3/ftztGIdSysVJnqq1fmrJc1fWWh9sWPfcJNcm+Vyt9Q9LKT+R5JuS/E6t9QuNGj3BSikvSfKSJH+u3wTgTFRKma21HiqllCSvTXJRkn1J3l1r/cp4W7e6Sq21nULdBBWdvlBdnGSWUv5tkm9Icl+Sv0pyXpJrknyl1vo/NKj7F0k+k0FAcVeS99Za/3Kl9U5Sv4vXr5OT7VLKLUmel+RPk7wyyd8keSrJH9Za396kzSdZV6cfwkopr6m13tPg7/9FkvfVWh9rsVmnWlcr26KUcrIRfiXJvbXWv9u0/gnr+qVa6//WsEZJ8gMZ7GMfqrV+dfj8a2utd7fQzJOts1GA1VXI22WfXEr5DxnsB19Ksi7JwSTHkqyrtf50w9qth0GllAeSLB5Ay/DnxUk+WWv97oa1X7DYR5ZSXp3hh/MMtnM7B+1nrq/p/nZrkvVJnkwyk+Qf1lqPllI+Wmt9VcO23ZXkjzM4Nr0syQeSzCe5odb6fU1qL1lHawHI8L3x6jzzS4X3txGslFLurbVeW0r5l0m+N8n7k7wiyYFa6//csHZXoVtXdV9aa/2TUsrzkrwhyYsz6Ive0laY13Uw1ub5RVdfZHUVpHdxzrmkbhft7eqY2tV58jcn+fEMtsN7kvyPGXwe+fVa674W2r1qX5z29XxoWPsbMtjP/iyD/ucfJvlikrfXWv+mQd2u9otzkvy3OeG9l+T3WjiH66oP+mit9VWllF/NYNt+NMmlSS6vtf5Yw9qdB0Ktnl+0cf7XYVDR5QvVyUlmKeUPTnbCfqrnl1H3/lrrK0sp5yf54SSvyeDDzd211l9fad1h7a5ev05OtkspH6m1fu+S3z9ca/27pZT7aq3XNKjb5YewC072dJLfqrVe1aDuYxmEVxuS3JvkPbXWXSutt6Rul9viiQwOEuWEdWyptX5Lg7qfTfLZJF9Ni20upfx2Bp34VzJ4X/zjWuunW/pA2kmA1VXI23Gf/LFa6/cMH++qtV4yfHx/rfWVDep2EgaVUn42ybdn8B7eOnzu92ut37/SmktqL27nf5NB/3l3Bh/ON9Vab2xQt6v97enjWyllS5J/l+TnkvxyC++Rp1//Usona60vOfH5FdbtJAAppfxmkkcz6JdflWQ6yeeTfKnWestK6w5rL+4XH0vyyiUB8sdrrd/VoG5X50NdhnmL2+JtST6Rr/VF/6DW+gMN6na1X3R5TO3ki6wO+86uzjm7DP67OqZ2dZ78oSS/Naz7hiS/mEGA8+Za69UN29zV/jZR50PD2r+X5JEk52SwLX4vg33u+2qtr2tQt6v94h1Jdib5SJ753vv2WutPNKjb2Zfpi5/tTvyM1/QcYFijk3Pazr5gqbU2/pfkD5bz/DLq3rf055Ln72+zzUm2JNma5PIkH21Y91eS3JbkR5P8veHP30jy/zSs+3X/zxmcDP10j1+/+5c8/mRbr1+Sdyb5+STfn+SWJL/aUt2fzeAgd/WS536/6fYd1jmW5D8muf2Efwfa2MZJnp/kR5L8dpKHMviw1Ndt8XCSbzrJ8x9uWPeHk/xOkhuTrGmrzUm2Lnn8t5N8KMkPNe0rhvWeyOAgcf/w5+Ljz7W0X5yf5OZh//bBJP+0Yd0u++RtSx6/5mTbf4V1P7bk8a6W2/zcJP99kv803Cfaeo989MS2t7QtutrftiV57pLf/1YGJylHWtgW70vyC0n+TZKPDffnGzM4kW9lGyd5zpLnP96w7v0n/P6R4c9G/duwxuEkb09yIMnzljz/UMO6XZ0PdVJ3cbtm8IHugxl+Abn4evZ0v+jymPqRE37/8PDnfQ3rdtJ3prtzzi77+vuHP9s+pt6/5HGb58lLt8XulrdFV/vbRJ0PneT1++NTbaMe7RcPLOf5ce8Twxo/meQ3M/jM9NtJ/kmSX0vyf7VQu5Nz2q6OI23NafNQKeW2JB/O4MPpeRkkS480rPu24bdW/2X4bffHMjjwNx5CmeScUspza61frrXuLKVcl8HOcHGTorXWf1VKeWmS70jy32SQZL611vonDdv7dd/Q1VqPJHlrw7pJd6/fX5dSfiHJVJLPlVJuTvKXGXwL0sRPJLkuySUZfMO2eHnRjzcpWmv9t6WU5yb5R6WUNyS5o1Ern+mTSX6+1np06ZOllDvbKF4Hwz5/N8nvllLWZPANb5N6XW6LV2eQaJ+o0QiFWut7krynlPL9Sd5RSvlEknOb1Bx6TillutZ6vNb6F2VwucpbM/j2o6k9Sa6rtf7V0idLKR9uoXbqYBj0rUluLaWsz2D4ZxNvK6X8v+mmT/7pUso5tdan6vCSweE++CsN6y49zv3rJY/LiQsuV631y0l+Y7hNfjKDb5nacFkp5Q+SzJXhpVLDbyGnG9btan/72Qy+EXw8SWqtfzkckr7ibxqXeF0Gw8T/PMkvJfmpJN+Y5PqGdedKKW9P8m0ZfOu/2Cd9Y8O6u0opv5HBt5lXZ/ChI2lnDsErhz//1wxG/qWUMjX8vYlOzodOUveHk7yjhbrJIMT7zxkM799aSvl4BkPb39Ow7tzw2+hW94uOj6mPl1J+PoN97nuS7B4+f07Dul31nV2dc3bW1y/q4Ji6eJ78X+Vr58mfT/Pz5M8OPz+dk0Gf9GvDuvMN6ybP3N+uTnv7254kP1xPuPynx+dDyeCLm0X/dMnjptuiq89P7y2lvC+D4Grxvfc9Sd7bsO6J+8Se4fNNt0Nqre8opXwkyfdlMFhhTZLfrLW2cb51qpyh6eWwnZxftDmnzUszOKF4QQZBxUyt9X9voe7fzuCFWjes+y211v+zhbpXJNlfa318yXNrkvzrWusvNa0/aYbb41UZfMj9SpJamw/lfl4GJ9v/3/DfTw3rv63WeqxZi7tRvnY99OeT/Iskr6u1vryFumuSvCjtz6H07Rm8Xq3PGbCk1poMPpD+cq11bRs1u1KWzJ9QSnllBteRbm+4jTcn2ZjBtz1Lt/F31Fr/qGF7Z4d1v3zC802v4f579YTr+Et7cxJ9dwbDPM/JYPjrU7XWX25atyvDfeJTtdanljz33CRv7HNfXwbXQT9Va90z/P35GVxCuOJ9bng8PT+D6++fnvMhg2+Vnmze6slRSvnWJb8eqrV+eRiA/Kum+0Up5bsyGCJ+d6115/C576y1/mGTul0ZHv/PzTP3i69kMCJyxeHCKc6zfibJ0Vpr4y8shu+RqzL4MPNXGZxof1vD/n7pfvEXtdYnSyn/LIP5CO5t1OCvrWPxmPqiJL/bpK8f1jsngy+yLkjy6ST31Fq/Wkr57lrrHzSo+5IMPnAsnSPmq0lSG8zFN6y9+OXm4meGT2QwQrbJcW8uyadP6Ov/eQb7YNP2fl+t9YMneb7psXrxPPnPM3j//VQGIdNDDeuWDN4bNcnHM/gcVZIcq7VuW2ndYe01Sf5lBpfD7M5gn/jrDEY3rXgOk1LKhiSfX3o+NHzvvWXpa7qCutcmOZpu5jr6lgxGYD+5eH5YBvPRfEet9UBRhJ8AAAVNSURBVIEGdZ+f5GcyeF/8aQaB0BeS/KcTQ60V1P7uJHPDescy6DcvaNhvLvZB52fYBw3b/J4m+8RqWJIzrM+gL/rDDEYON3n/bc7gkuDnJPmTDObFXPystuLjSFtz2ixeq7s0zZ5L8mhtNpdEJ3WHtU91u/MP1pYnQ+27MrgOOEm+nHavW156DXcyeB1bef26ULq9Hnqi5gw4xWvXyvX3XZm0bTys3dU13J3Mn9BVX9GlrrZxl4b73LoMDvJt7sud9XGTpsP3XievXZe62i9O0Q+1dQ7X1Xukq76zs36ow325q/1iNY57izXbmjeoqzZPXD+0yv1FG++9/zh8OEnn9l1t4677zaWf13v9mSHpvO9MWj5XbuvyqPekmwkZu6qbJAs5xWSoLdSeNP91feYEoD8yfHz/s//ZaXX5+nWhq+2QJH+nPnOizneVUn6ux3Un7bVLJm8bJ931Q129fl2+R7qydBsnXzux6HNf39U+N4mvX1e6eu912V90ZRLPASbt2NdlP7RYe6k2ane1X3TV3kn8zLB0WyzW7ns/NGn9xbdN4Ll9V9t40vrNrk1U39lKaFM7ula3q7pDnc4lMWE6uQ6449evC11eD71acwa0NTfTpL12yYRt46FO+qEOX7/O5wzowCT29V3tc5P4+nWlq/2iy/6iK5N4DjBpx74u+6GuanfVX0zacS/pbhtPYj80af3FJJ7bd9XmSes3uzZZfWdtOPPyif+GDb0xyS19rptkNkvudrF0PW1vk77/y+DNes4Jzz03yQ/19fWbtO2Q5IoMhsUtfe6cJP9dH+tO2ms3qdt4NfqhNl+/1egrOtgvJq6v73BfnrjXb9L2i9XokzvYFhN3DjBpx74u+6EO9+VO9otJO+51vI0nrh+atP6iy/ZO2rF60vrNrv9NWt/Z2kTEAAAAALTnVJPxAgAAADBGQhsAAACAHhLaAAC9UUp5qpSyo5TyyVLKu0opz3+WZa8upXznkt/fUEr5+ytc7+ZSyg0r+VsAgK4IbQCAPvlirfXSWutLknw5yRueZdmrkzwd2tRa31JrffsK17s5idAGAOgVExEDAL1RSlmotU4NH78hyZYkv5/kFzK4A8Pnkvx4kucl+aMkTyU5muSfJ/neJAu11v+7lPJtSf59krVJnkjyT2qtnyql/FaSY0kuT7Ihyf9Ua313KeWPklyUZF+StyX5UJLbh+t8TpIfqbX+WfdbAADga9acfhEAgNVVSlmT5PuT3Jvk40m+o9ZaSyn/OIOg5eZSylsyDGmGf/O9S0q8Nckbaq1/Vkq5MsmvJ3nV8L/NJvmuJC9O8t4k707yxiQ/V2t99bDWryX51Vrr75RSnpvBrVEBAFaV0AYA6JPnlVJ2DB8/kOQ/JHlRkjtLKbMZjHzZ92wFSilTGVw29a5SyuLT37Bkkd+rtX41ye5SyvpTlPlEkv+llLIpyXuMsgEAxkFoAwD0yRdrrZcufWI46uVXaq3vLaVcneQXT1PjOUm+cGKdJb60tPzJFqi13lFKeTDJDyb5QCnlplrrR0f5HwAAaIuJiAGAvvumJAeHj39qyfPHk0yfuHCt9ViSfaWU1yVJGfj206zjGbVKKRckeazW+u+S3J3B3DoAAKtKaAMA9N0vZnCp08NJ5pc8f0+S64a3CL/qhL/58ST/qJTyp0keTfLa06xjZ5KnSil/Wkr52SQ/luSTw0u1XpJkpXelAgBYMXePAgAAAOghI20AAAAAekhoAwAAANBDQhsAAACAHhLaAAAAAPSQ0AYAAACgh4Q2AAAAAD0ktAEAAADoIaENAAAAQA/9/5AUZ8Bgp/DYAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 1440x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from scipy.cluster.hierarchy import dendrogram, linkage\n",
    "\n",
    "plt.figure(figsize=(20,5))\n",
    "linkage_data = linkage(ml_df, method='ward', metric='euclidean')\n",
    "dendrogram(linkage_data)\n",
    "plt.xlabel('Patients')\n",
    "plt.yticks([])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "tags": []
   },
   "source": [
    "## References <a id=\"references\"></a>\n",
    "[1]  <a id=\"r1\"></a> Previous Jupyter Notebook: https://github.com/microsoft/genomicsnotebook/blob/main/sample-notebooks/fhir_long_read.ipynb  \n",
    "[2]  <a id=\"r2\"></a> Microsoft Blog Post:https://techcommunity.microsoft.com/t5/healthcare-and-life-sciences/convert-synthetic-fhir-and-pacbio-vcf-data-to-parquet-and/ba-p/3577038  \n",
    "[3]  <a id=\"r3\"></a> FHIR HL7: http://hl7.org/fhir/index.html  \n",
    "[4] <a id=\"r4\"></a> Parquet File Format: https://spark.apache.org/docs/latest/sql-data-sources-parquet.html  \n",
    "[5] <a id=\"r5\"></a> Python `pandas` library: https://pandas.pydata.org/  \n",
    "[6] <a id=\"r6\"></a> Broad GATK GVCF Explanation: https://github.com/broadinstitute/gatk-docs/blob/master/gatk3-faqs/What_is_a_GVCF_and_how_is_it_different_from_a_'regular'_VCF%3F.md  \n",
    "[7] <a id=\"r7\"></a> `bcftools` Documentation: https://samtools.github.io/bcftools/bcftools.html  \n",
    "[8] <a id=\"r8\"></a> HTSlib/SAMtools/Tabix Homepage: https://www.htslib.org/  \n",
    "[9] <a id=\"r9\"></a> Linkage Disequilibrium: https://en.wikipedia.org/wiki/Linkage_disequilibrium  \n",
    "[10] <a id=\"r10\"></a> Scikit-Learn Clustering: https://scikit-learn.org/stable/modules/clustering.html#clustering"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# NOTICES"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "THIS NOTEBOOK JUST PROVIDE A SAMPLE CODES FOR EDUCATIONAL PURPOSES. MICROSOFT DOES NOT CLAIM ANY OWNERSHIP ON THESE CODES AND LIBRARIES. MICROSOFT PROVIDES THIS NOTEBOOK AND SAMPLE USE OF  LIBRARIES ON AN “AS IS” BASIS. DATA OR ANY MATERIAL ON THIS NOTEBOOK. MICROSOFT MAKES NO WARRANTIES, EXPRESS OR IMPLIED, GUARANTEES OR CONDITIONS WITH RESPECT TO YOUR USE OF THIS NOTEBOOK. TO THE EXTENT PERMITTED UNDER YOUR LOCAL LAW, MICROSOFT DISCLAIMS ALL LIABILITY FOR ANY DAMAGES OR LOSSES, INCLUDING DIRECT, CONSEQUENTIAL, SPECIAL, INDIRECT, INCIDENTAL OR PUNITIVE, RESULTING FROM YOUR USE OF THIS NOTEBOOK.\n",
    "\n",
    "#### Notebook prepared by [Tim Dunn](https://github.com/TimD1)- Research Intern- Microsoft Biomedical Platforms and Genomics"
   ]
  }
 ],
 "metadata": {
  "kernel_info": {
   "name": "python38-azureml"
  },
  "kernelspec": {
   "display_name": "Python 3.8 - AzureML",
   "language": "python",
   "name": "python38-azureml"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  },
  "nteract": {
   "version": "nteract-front-end@1.0.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}