1A.[Omics_revamped]GenerateLLM_analysis.ipynb

{
 "cells": [
  {
   "cell_type": "raw",
   "metadata": {
    "tags": []
   },
   "source": [
    "This notebook queries GPT-4 and formats its response to obtain a succint name for each of the 'omics gene sets "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import pandas as pd\n",
    "import json \n",
    "import openai"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import math"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.openai_query import openai_chat\n",
    "from utils.prompt_factory import make_user_prompt_with_score\n",
    "from utils.llm_analysis_utils import process_analysis, save_progress"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "openai.api_key = os.environ[\"OPENAI_API_KEY\"] # Environment variable"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "runVersion = \"additional\"; # initial"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "geneSep = \" \"\n",
    "inputFilePath = \"data/omics_revamped_LLM_DF.tsv\"; #\"data/omics_revamped.txt\"\n",
    "jsonFilePath = \"jsonFiles/OmicsRunLLM.json\"\n",
    "genesCol = \"GeneList\"\n",
    "nameCol  = \"GeneSetName\"\n",
    "outputFilePath = \"data/omics_revamped_LLM_DF.tsv\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(jsonFilePath) as json_file:\n",
    "    config = json.load(json_file)\n",
    "    \n",
    "context = config['CONTEXT']\n",
    "gpt_model = config['GPT_MODEL']\n",
    "temperature = config['TEMP']\n",
    "max_tokens = config['MAX_TOKENS']\n",
    "rate_per_token = config['RATE_PER_TOKEN']\n",
    "LOG_FILE = config['LOG_NAME'] + '240129'+'log.json'\n",
    "DOLLAR_LIMIT = config['DOLLAR_LIMIT']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "SEED = 42"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'gpt-4-1106-preview'"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gpt_model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Run GPT-4 query pipeline for NeST gene sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "df = pd.read_csv(inputFilePath, sep = \"\\t\"); "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "if runVersion == \"initial\":\n",
    "    df['LLM Name'] = None\n",
    "    df['LLM Analysis'] = None\n",
    "    df['Score'] = None"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1641\n",
      "1564\n",
      "1576\n",
      "1328\n",
      "1594\n",
      "1565\n",
      "1581\n",
      "1550\n",
      "1682\n",
      "1726\n",
      "71\n",
      "1847\n",
      "1644\n",
      "1603\n",
      "1544\n",
      "1532\n",
      "1471\n",
      "1853\n",
      "1649\n",
      "1407\n",
      "1694\n",
      "81\n",
      "1735\n",
      "1781\n",
      "1688\n",
      "1628\n",
      "1685\n",
      "1456\n",
      "1628\n",
      "1685\n",
      "1421\n",
      "1523\n",
      "91\n",
      "1617\n",
      "1618\n",
      "1839\n",
      "1598\n",
      "1535\n",
      "1357\n",
      "1552\n",
      "1896\n",
      "1575\n",
      "1729\n",
      "101\n",
      "1547\n",
      "1779\n",
      "1326\n",
      "1383\n",
      "1670\n",
      "1617\n",
      "1806\n",
      "1679\n",
      "1850\n",
      "1767\n",
      "111\n",
      "1569\n",
      "1570\n",
      "1551\n",
      "1774\n",
      "1634\n",
      "1643\n",
      "1458\n",
      "1458\n",
      "1655\n",
      "1438\n",
      "121\n",
      "1621\n",
      "1513\n",
      "1381\n",
      "1908\n",
      "1700\n",
      "1767\n",
      "1532\n",
      "1550\n",
      "1722\n",
      "1736\n",
      "131\n",
      "1824\n",
      "1753\n",
      "1573\n",
      "1660\n",
      "1611\n",
      "1863\n",
      "1768\n",
      "1559\n",
      "1609\n",
      "1888\n",
      "141\n",
      "1640\n",
      "1662\n",
      "1894\n",
      "1610\n",
      "1771\n",
      "1781\n",
      "1708\n",
      "1538\n",
      "1711\n",
      "1720\n",
      "151\n",
      "1698\n",
      "1794\n",
      "1846\n",
      "2007\n",
      "1675\n",
      "1757\n",
      "1546\n",
      "1737\n",
      "1629\n",
      "1630\n",
      "161\n",
      "1769\n",
      "1643\n",
      "1760\n",
      "2009\n",
      "1760\n",
      "1782\n",
      "1749\n",
      "1621\n",
      "1886\n",
      "1606\n",
      "171\n",
      "1544\n",
      "1632\n",
      "1651\n",
      "1638\n",
      "1834\n",
      "1645\n",
      "1839\n",
      "1747\n",
      "1877\n",
      "1936\n",
      "181\n",
      "1697\n",
      "1654\n",
      "1526\n",
      "1742\n",
      "1629\n",
      "1735\n",
      "1664\n",
      "1542\n",
      "1663\n",
      "1702\n",
      "191\n",
      "1705\n",
      "1909\n",
      "1684\n",
      "1561\n",
      "1657\n",
      "1744\n",
      "1685\n",
      "1701\n",
      "1829\n",
      "2070\n",
      "201\n",
      "1504\n",
      "1586\n",
      "1843\n",
      "1623\n",
      "1694\n",
      "1805\n",
      "1676\n",
      "1604\n",
      "1743\n",
      "1708\n",
      "211\n",
      "1904\n",
      "1810\n",
      "1625\n",
      "1983\n",
      "1680\n",
      "1597\n",
      "1631\n",
      "1553\n",
      "1765\n",
      "1795\n",
      "221\n",
      "1709\n",
      "1685\n",
      "1486\n",
      "1612\n",
      "1549\n",
      "1837\n",
      "1487\n",
      "1874\n",
      "1665\n",
      "1648\n",
      "231\n",
      "1908\n",
      "1877\n",
      "1543\n",
      "1743\n",
      "1559\n",
      "1647\n",
      "1764\n",
      "1723\n",
      "1888\n",
      "1937\n",
      "241\n",
      "1816\n",
      "1876\n",
      "1700\n",
      "1606\n",
      "1817\n",
      "1688\n",
      "1620\n",
      "1745\n",
      "1675\n",
      "1869\n",
      "251\n",
      "1821\n",
      "1534\n",
      "1889\n",
      "1792\n",
      "1888\n",
      "1633\n",
      "1892\n",
      "1673\n",
      "1839\n",
      "1507\n",
      "261\n",
      "1428\n",
      "1763\n",
      "1765\n",
      "1762\n",
      "1831\n",
      "1693\n",
      "1635\n",
      "1743\n",
      "1820\n",
      "1795\n",
      "271\n",
      "2079\n",
      "1720\n",
      "1555\n",
      "1392\n",
      "1403\n",
      "1750\n",
      "1544\n",
      "1962\n",
      "1681\n",
      "1644\n",
      "281\n",
      "1660\n",
      "1799\n",
      "1793\n",
      "1553\n",
      "1554\n",
      "1538\n",
      "1597\n",
      "1670\n",
      "1590\n",
      "1862\n",
      "291\n",
      "1808\n",
      "1625\n",
      "1840\n",
      "1634\n",
      "1646\n",
      "1647\n",
      "1777\n",
      "1700\n"
     ]
    }
   ],
   "source": [
    "for i, row in df.iterrows():\n",
    "    \n",
    "    term_genes = row[genesCol]\n",
    "    genes = term_genes.split(geneSep) \n",
    "    \n",
    "    if runVersion == \"additional\":\n",
    "        if type(row['LLM Name']) == str:\n",
    "            continue # skip this row because already done\n",
    "   \n",
    "    \n",
    "    prompt = make_user_prompt_with_score(genes)\n",
    "\n",
    "    analysis, finger_print = openai_chat(context, prompt, gpt_model, temperature, max_tokens, rate_per_token, LOG_FILE, DOLLAR_LIMIT, SEED)\n",
    "\n",
    "    if analysis:\n",
    "        llm_name, llm_score, llm_analysis = process_analysis(analysis)\n",
    "        df.loc[i, 'LLM Name'] = llm_name\n",
    "        df.loc[i, 'LLM Analysis'] = llm_analysis\n",
    "        df.loc[i, 'Score'] = float(llm_score)\n",
    "\n",
    "    else:\n",
    "        #go_term = row['GO']\n",
    "        name = row[nameCol]\n",
    "        print(f'No analysis for {name}')\n",
    "        df.loc[i, 'LLM Name'] = None\n",
    "        df.loc[i, 'LLM Analysis'] = None\n",
    "    #if (i%10 ==1):\n",
    "    #    break\n",
    "        \n",
    "    # Keep on saving to not loose data if something happens\n",
    "    if (i%10 == 1):\n",
    "        print(i)\n",
    "        df.to_csv(outputFilePath, sep = \"\\t\",  index=False)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.to_csv(outputFilePath, sep= '\\t', index=False)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "llm_eval",
   "language": "python",
   "name": "llm_eval"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}