GNN_practice.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "13a42f7a",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using backend: pytorch[12:56:19] /opt/dgl/src/runtime/tensordispatch.cc\n",
      ":43: TensorDispatcher: dlopen failed: libtorch_cuda_cu.so: cannot open shared object file: No such file or directory\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import torch\n",
    "import dgl\n",
    "\n",
    "from rdkit import Chem\n",
    "from rdkit.Chem import AllChem\n",
    "\n",
    "from tdc.single_pred import ADME"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "52603523",
   "metadata": {},
   "source": [
    "# Molecular Graph Practice"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b42910b6",
   "metadata": {},
   "source": [
    "# 1. Download datasets from therapeutic Data Commonds (TDC)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "51081704",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Found local copy...\n",
      "Loading...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "dataset = ADME(name='Solubility_AqSolDB')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "f8cfe39a",
   "metadata": {},
   "outputs": [],
   "source": [
    "split = dataset.get_split(\n",
    "    method='random',\n",
    "    seed=999,\n",
    "    frac=[0.7, 0.1, 0.2],\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "5b0591c9",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_set = split['train']\n",
    "valid_set = split['valid']\n",
    "test_set = split['test']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "8d2396dc",
   "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>Drug_ID</th>\n",
       "      <th>Drug</th>\n",
       "      <th>Y</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>N,N,N-trimethyloctadecan-1-aminium bromide</td>\n",
       "      <td>CCCCCCCCCCCCCCCCCC[N+](C)(C)C.[Br-]</td>\n",
       "      <td>-3.616127</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>4-chlorobenzaldehyde</td>\n",
       "      <td>O=Cc1ccc(Cl)cc1</td>\n",
       "      <td>-2.177078</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>zinc bis[2-hydroxy-3,5-bis(1-phenylethyl)benzo...</td>\n",
       "      <td>CC(c1ccccc1)c1cc(C(=O)[O-])c(O)c(C(C)c2ccccc2)...</td>\n",
       "      <td>-3.924409</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4-({4-[bis(oxiran-2-ylmethyl)amino]phenyl}meth...</td>\n",
       "      <td>c1cc(N(CC2CO2)CC2CO2)ccc1Cc1ccc(N(CC2CO2)CC2CO...</td>\n",
       "      <td>-4.662065</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>vinyltoluene</td>\n",
       "      <td>C=Cc1cccc(C)c1</td>\n",
       "      <td>-3.123150</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6983</th>\n",
       "      <td>sulfamethazine</td>\n",
       "      <td>Cc1cc(C)nc(NS(=O)(=O)c2ccc(N)cc2)n1</td>\n",
       "      <td>-2.730000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6984</th>\n",
       "      <td>sulfasalazine</td>\n",
       "      <td>O=C(O)c1cc(/N=N/c2ccc(S(=O)(=O)Nc3ccccn3)cc2)c...</td>\n",
       "      <td>-6.140000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6985</th>\n",
       "      <td>tetracaine</td>\n",
       "      <td>CCCCNc1ccc(C(=O)OCCN(C)C)cc1</td>\n",
       "      <td>-3.010000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6986</th>\n",
       "      <td>tetracycline</td>\n",
       "      <td>CN(C)[C@@H]1C(O)=C(C(N)=O)C(=O)[C@@]2(O)C(O)=C...</td>\n",
       "      <td>-2.930000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6987</th>\n",
       "      <td>verapamil</td>\n",
       "      <td>COc1ccc(CCN(C)CCCC(C#N)(c2ccc(OC)c(OC)c2)C(C)C...</td>\n",
       "      <td>-3.980000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>6988 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                Drug_ID  \\\n",
       "0            N,N,N-trimethyloctadecan-1-aminium bromide   \n",
       "1                                  4-chlorobenzaldehyde   \n",
       "2     zinc bis[2-hydroxy-3,5-bis(1-phenylethyl)benzo...   \n",
       "3     4-({4-[bis(oxiran-2-ylmethyl)amino]phenyl}meth...   \n",
       "4                                          vinyltoluene   \n",
       "...                                                 ...   \n",
       "6983                                     sulfamethazine   \n",
       "6984                                      sulfasalazine   \n",
       "6985                                         tetracaine   \n",
       "6986                                       tetracycline   \n",
       "6987                                          verapamil   \n",
       "\n",
       "                                                   Drug         Y  \n",
       "0                   CCCCCCCCCCCCCCCCCC[N+](C)(C)C.[Br-] -3.616127  \n",
       "1                                       O=Cc1ccc(Cl)cc1 -2.177078  \n",
       "2     CC(c1ccccc1)c1cc(C(=O)[O-])c(O)c(C(C)c2ccccc2)... -3.924409  \n",
       "3     c1cc(N(CC2CO2)CC2CO2)ccc1Cc1ccc(N(CC2CO2)CC2CO... -4.662065  \n",
       "4                                        C=Cc1cccc(C)c1 -3.123150  \n",
       "...                                                 ...       ...  \n",
       "6983                Cc1cc(C)nc(NS(=O)(=O)c2ccc(N)cc2)n1 -2.730000  \n",
       "6984  O=C(O)c1cc(/N=N/c2ccc(S(=O)(=O)Nc3ccccn3)cc2)c... -6.140000  \n",
       "6985                       CCCCNc1ccc(C(=O)OCCN(C)C)cc1 -3.010000  \n",
       "6986  CN(C)[C@@H]1C(O)=C(C(N)=O)C(=O)[C@@]2(O)C(O)=C... -2.930000  \n",
       "6987  COc1ccc(CCN(C)CCCC(C#N)(c2ccc(OC)c(OC)c2)C(C)C... -3.980000  \n",
       "\n",
       "[6988 rows x 3 columns]"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "c4ea5457",
   "metadata": {},
   "outputs": [],
   "source": [
    "smi_train = list(train_set['Drug'])\n",
    "label_train = list(train_set['Y'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "f8144afe",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'CCCCCCCCCCCCCCCCCC[N+](C)(C)C.[Br-]'"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "smi_train[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "3277eff0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "-3.6161271205"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "label_train[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "64bea007",
   "metadata": {},
   "source": [
    "# 2. Visualize molecules"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "6b604de2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<rdkit.Chem.rdchem.Mol at 0x154b8ea7e710>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mol = Chem.MolFromSmiles(smi_train[0])\n",
    "mol"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "8fc458a0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAcIAAACWCAIAAADCEh9HAAAABmJLR0QA/wD/AP+gvaeTAAAgAElEQVR4nO3deVzU1f4/8NewrwoKyI4yIG6hlmkG5UaYiqnfRK2uWS5j3ZvWTW/cvlZm39+ladHrzcxG0bQsEcKSSCu8WLnlgruIArHIKsgOwzLM+f1xdCRcQmY+n8/M8H7+5Xyc+Zw3D+U15/M553OOjDEGQgghXWUhdQGEEGLaKEYJIUQvFKOEEKIXilFCCNELxSghhOiFYrS7q6/H+fPX/9zSgkuXJK2GEBNEMdrdnTqF0FAcPAgAV69iwQKpCyLE1FCMEowbh7//Ha2tUtdBiGmiGCUICMCkSVizRuo6CDFNVlIXQIzC66/jwQfx6KNS10GICaLeaHd06/W7vT3efRevvy5FNYSYOIrR7igsDNOmobj4DwenTkXPnhIVRIgpo4v6buf4cRw/Djc39OoFAM7OiIq6/leffIL0dAlLI8QkUW+021GpAOD552FnBwCbNmHOHGzeDAA+PnjiCSlrI8QUyWihvG6lpgY+PmhsRGYm+vdHQwO8vVFbi3PnMGQI1Gp88gmefRbu7lIXSojpoN5o9/LFF2howIQJ6N8fAL76CrW1eOQRDBkCAImJWL4cM2ZIWyMhJoZitHvZuBEAFi++/pJf4Hd4SQ8yEXJP6KK+Gzl0COHh8PREQQGsrXH6NIYPR+/eKCyEnR3OnUNoKHr2RFERHB2lrrVz1Gq1vb291FWQ7o56o93Id9/93Ldvy/z5sLYGgPXrgXZjTbwr+uyzppGh27Ztc3R0fO2116QuhBCa8NRtVFZW/uc/k1pbNQpFIdCnrg47d0Imw6JFANDYiC+/BACFQtoyO8vFxaWxsTEnJ0fqQgih3qhxYIx9/PHHy5YtE+4ey9atW5uamiZOjAwI6APgiy9QV4fx42+ONVVXIzz8+liT8QsKCgKQnZ0tdSGEUIwaB5lMtmrVqjVr1pSUlAjUxObNmwEsvjGcdODAOUfHm33PuDiGdmNNxk8ul1tYWOTm5rbSylREahSjxkLQ7tX+/fszMjJ8fX2nTJkC4MiRI/HxoV5eI6ZPZwDS09Pz8h6YPPnQzJlCNC4IOzs7Hx8fjUZTUFAgdS2ku6MYNRaCxqhKpQKwcOFCS0tL3cvo6EgbGxl/WVZ2auDAb/hYk6mg63piJChGjYVcLgcgxJhJRUXFt99+a2VltXDhQgDV1dWJiYkymWz+/PkA6urq4uPjASwwtfmiFKPESFCMGgseo0KEwpYtW5qbm6dMmeLj4wNg69atjY2Njz/+OI+h7du319XVjR8/fuDAgQZvWlDCffEQck9owpOx4KEmRCi8+OKLPXv2HHJjDD4uLg7txpo2bdrU/qUJod4oMRL0FJOxKC8v9/Dw6NGjR01NjXCt/PLLL2PHjvXy8srPz7e2tv7tt99Gjx7t5uZWWFhoa2srXLtCOHv2kkLxz4CA8J07l0ldC+nWqDdqLNzd3V1cXKqrq8vLy90FW2GJDy4pFApra2u0G3oyuQwFEBgYcuzYN6dPQ6uFBd2dItKh/31GRLjbo1xFRcWuXbssLS2ff/55fsTGxsbe3p4PPZkcJyf06YPmZhQWSl0K6d4oRo2I0Df7EhMTQ0NDo6KiAgIC+BGVSlVWVsbj2xQFBQEA3R0l0qIYNSLCjTIBYIytWbPm+PHjHUaTnJ2dhWhOHBSjxBhQjBoRQS/q9+3bl52d7e/vHxkZKcT5JcG70TTliUiLYtSIiPAgk0Kh4A8ymQfqjRJjQDFqRISbT15aWpqcnGxlZaUbXDIPPEapN0qkRTFqRLy8vBwdHSsqKqqqqgx75ri4uNbW1unTp3t7exv2zNLS9UZp9jOREMWoEZHJZEJ0SLVabYdV8syGiwt690ZDA0pLpS6FdGMUo8ZFiNuje/fuzcvLk8vl48ePN+BpjcGlS9i8GSdPondv5OWhvl7qgki3RDFqXISIUT64tHjxYguze9Zn4ULExWH4cNjY4K23cOqU1AWZgtOnT2/evDkrK0vqQsyHuf1emTqDX9QXFhbu2bPHxsZm3rx5hjqnUWEM33wjdREmZf369QsXLkxNTZW6EPNBMWpcDN4b3bRpU1tb28yZMz08PAx1TqPyr3/hzTdRVyd1HaaDFhg0OIpR42LYGNVoNOY6uKTj6YlFi7Bq1fWXW7agslLSgoweLTBocBSjxsXX19fOzq60tLS8vFz/s6WkpBQVFQ0YMOCRRx7R/2xGRa2GWn39z3/7Gw4cQHY2zp7FggXw9sasWdi3T9L6jBjFqMFRjBqXHTt2yGQyDw+P4ODgxYsXnz9/Xp+z8cGlF154QSaTGahAo1BQgDFjoFuXysoKH3+Mo0dhY4NJk9DaisREPPYYQkOxfj2EXL7VJAUFBclkst9//12r1Updi7lgxDg0Nzf/9a9/5f8ouknyMpls3Lhx8fHxzc3N93rC/Px8S0tLe3v7a9euCVGwVFJSmKsrA1hwMNu+nTU2Xj++axcrKWGMscJCplQyPz8GMIDZ2bHoaJaaKmHJRqdPnz4ACgoKpC7ETFCMGoXCwsLRo0cDsLW1Xbt2LWPswoULS5cudXJy4nnq6uq6dOnS7Ozszp/z9ddfBzBv3jyhir6FVqutqqoS8vxMqWQWFgxgUVHs7k21tLDERDZhApPJrufpnDlbVSpVXV2dcBWairCwMABpaWlSF2ImKEal98svv3h6egLw9fX97bff2v9VbW2tSqUaPnw4D1MLC4uIiIiEhISWlpa7n7OlpcXLywvA4cOHhaz9utbW1piYmKeeesrW1jY6Ojo1NVWr1Rq2iWvX2OOPM4BZWrKVK1lbW2c/ePkyW76cDRzYam3tAMDZ2VmhUJw8edKw5ZkWPvtt48aNUhdiJihGpaTVateuXcv38xg7dmxpaemd3nnixAmFQuHo6Mjz1NPTMyYmJjc3907v37lzJ4DQ0FBB6v6j4uLi8PBwAFZWVrrlo0JCQtasWWOo+wmnTrHAQAaw3r3Zjz925QzNzc0JCQkRERG628QPPPCASqWqr683SIWm5Z133gEQExMjdSFmgmJUMrW1tdHR0fwGaExMjEaj+dOPVFdXq1Sq0NDQDp3T1tbWDu/kz32uX79emNpvOnDgAO/2+vj4HD58uLi4WKlU6lbXN0jn9MsvS+3sGMBGjmT5+foWnJGR8fLLL7u6uurulrz88suZmZn6ntek/L5r14GhQyvmzpW6EDNBMSqNzMzMQYMGAejRo8fXX399rx/nnVMHBwfdkFRMTEz+jYzJzs6WyWSOjo41NTWGLvwPVCoV70o/+uijJXx8hzHGWFtbW2pqanR0tJXV9T0TBwwYoFQq77Vz2tTUtHTpUnf3Pl5eV+bOvTmapL+mpibeOeXlLVmypH395u/YMQawoUOlrsNMUIxK4JtvvunZsycPl4yMjC6fp6qqSqVSDR48uEPn9NVXXwWwaNEiA9bcQV1d3ezZs3lXeunSpXe6V1tYWKhUKv38/HiFdnZ2vHPamSby8/MffPBBAPb29tu3f2fQ8m86ffr0kCFDAKxZs0agJoxRZSUDmJMTM/Qt7O6JYlRUfCiG3557+umnDXJjTqvV7tu3Lzo6mncMAfAblD927SZiJ1y+fJlHj7Ozc0JCwp++X6PRpKamRkVF6e5LDho0aO3atXcZNN+/fz+flOPv73/s2DGDlt/RmjVrAPztb38TtBWj06sXA1i36oMLhmJUPFevXp0wYQIfilEqlQY/f2lpqVKp9PDwsLOzA2BjYyPEoHlycrKLiwsfRDp//vw9fTYrKysmJsbd3Z2H6W0HzbVarVKp5N8EkyZNEmHSa3JyMoCJEycK3ZBxGTmSAezAAanrMAcUoyI5ePAgn1Tv4+Nz6NAh4Rriz30OHz5cd18yJCRk9erV+ueRRqNZuXIlX21v2rRp1dXVXTvPXQbNa2pqnnzySd2wW1vnpzXpISMjA4BcLhehLSPy9NMMYJ99JnUd5oBiVAwqlcrGxubWoRiDy8jIkMlkzs7OtbW1fNC8b9++Bhk0Ly8vf+yxx3RdaYP0cDsMmru4uPBlqHr16vX999/rf/5OampqsrCwsLKy+tPZuGblrbcYwN54Q+o6zAHFqLAaGxufe+45HhMKhULoX9SlS5cC+Otf/6o7YpBB8xMnTvA4dnd337dvn2Frbj9oPn78+GHDhuXk5Bi2iT/l7+8PICsrS+R2pbRtG3+0S+o6zAHFqIAuX7583333AXBycurMUIyeGhsbec/uto/odHnQXNeVDgsLKyoqEqDw606fPl1TU9NowGlNncan2e7du1f8piVz6BAD2IgRUtdhDihGBbRq1SoAAwcOvHjxogjNbdmyBcDo0aPv8p57GjRXq9ULFizQdaW7sDyKqVAoFADWrVsndSEiunqVyeUsOlrqOswBxaiA2traYmNja2trxWlu1KhRALZu3dqZN/NBc92S+LcOmreftvmZuQ9EvPfeewBeeeUVqQsRUX09W7uWvfQSW72aifVf1FxRjJqJM2fO8FGahoaGzn/qToPmSUlJ/P5AcHDw2bNnhSvbSCQlJQGIioqSuhCxtLaykSPZ+vXs/Hn26ads2DDW1CR1TSaMYtRMvPDCC/r0pzoMmvOZp/rMajIt/EtowIABUhcill272LPP3ny5aBHbvl26akyejDEGYuLq6+t9fHxqa2vPnz+veza0C9RqdUJCgkqlmjNnTkNDwz//+U8zWzb/ThobG52cnKytrRsbG3WLVJmz996DhQX+8Y/rL9evR1kZ3nlH0ppMmJXUBRAD+Oqrr2pra8eMGaNPhgKwt7efN2+euW7FfBcODg5eXl7FxcVXrlzRzbQ1Zy4uKCq6+bKmBjcuREgX0F5M5mDjxo0w6+0/RdCNNnqrrsa4cfj22+ubAjY34+uv8dhjUpdlwihGTd7x48fT09Pd3NxmzJghdS0mrFvEaFMTFizAQw/B0xMvv4ywMDz7LB5+GIsWYcgQqYszYXRRb/L49p/PP/88HxciXSOXywHk5ORIXYhgcnLw5JM4cwaOjte3on7uOZSUwMsL3eF2sJCoN2raampq4uPjZTLZQt12w6RLJOyNXrhw4cKFC59//nlTU5NQbXz/PR58EGfOIDgYR44gPBwALC3h60sZqj+KUdP2xRdfNDQ0TJgwoX///lLXYtp4b1T8GN2xY8eoUaNmzJgxb948T0/PxYsXnzt3zpANMIb33sMTT6CqClOn4tgx3HefIc9PQPvUm7hhw4YBSExMlLoQk1dTUwPA3t7e4Hua3klLSwtfSgbAmDFj+DNjAGQy2dixY3fs2NGk/5T4igo2cWJX9lMl94Ji1IQdPHgQgKenZ/da4U0wfD3pwsJCEdoqKysbN24c/riG94ULF2JiYtqvHKhQKC5cuNC1Jk6eOKEZPJgBzN2dGXpdLtIexagJmzt3LoD//d//lboQMzF69GgAP//8s9AN/frrr+23U+3wt7W1tSqV6v7779d1TvkWW/f0Zblt2zYHB4f/GzmSjRjB7rwRNzEIilFTVVVV5eDgYGFh8fvvv0tdi5ngX0txcXGCtnKn7VRvxfd/dXR05Hnq6ekZExPzp//carVaN974wgsvtNDD8sKjGDVVH374IYBJkyZJXYj5WLVqlaOjo5+fH9/RxODn/9PtVCdPnrx06dLs7Oz2B2tqalQq1dChQ3ky6vZ/bW1tvbWJgoICvtCXnZ2d0N8HRIdi1AQ0Nzfn5OSkpqaqVKqYmJjo6OhBgwZZWlpaWFiY/RJ2YmppadFtVeDq6vrKK6/os/11B5cuXdJtp3rbIcGLFy/qgjIyMjIpKalDUPLOqYODA3+bl5dXTExMXl6e7g1paWl85UMRtlMl7VGMGpeGhoYzZ84kJSW9//77CoVi/Pjx/v7+fBe5DviiIY888shteyWka/jiLHxHE46vHKjnmvy7d+/u2bMngJCQkLsMGXXmKr66ulqlUg258dAR75zu3LkzNjaWL6oyefJkEbZTJe1RjEpGrVafP38+ISFBqVQqFIqIiIjAwMDbJqaVlVVgYGBERIRCoVAqlQkJCSdOnMjNzfXx8QHw6quvSv2jmKGMjIxbB83vdUNpdmM7Vf6dN2fOnNvuMtABv4oPDQ29y1W8VqtNS0ubPXs239+Ff61aWFi8/fbb4mynStqjGJVAVlaWbq/2DmxsbEJCQqZMmfLKK6+sW7du7969WVlZdxqiPXLkCP8t2rlzp8g/QjehVqu3bdumGzQHEBYW1vlB8/Lyct6xbT+rqfM6XMV7e3vHxMTk5+e3f09ZWZlSqXRxcbG0tFy9evW9NkEMgmJUbAcPHnRwcLCysrKxsWnfx0xOTs7JydFoNPd0to8++giAk5NTFzpKpPN4ojk5OXV+0Pz48eMBAQEA3N3d//vf/3a56aqqKpVKpVsCUdc5bf9fhY9cff75511uheiDYlRsfFZNJ5epb2xsPHv27K5duz744IPFixdPmDChb9++HcY9+PKg/fv37yYr1Uuo84Pmuu1Uw8PDi4uL9W9aq9Xu27cvOjpadxXfPjRXrFgBYOXKlfo3RLqAYlRUlZWVfLJn7i0zotVqdU5OTnJy8p/eKt29e3f7DzY2Ng4fPhzAtGnTRHuQsZu7y6C5Wq2eP38+P65QKAz+gFlpaWlsbOzQoUPb32blm8I+88wzhm2LdBLFqKhWr14NYMqUKbojbW1t4eHhuh06b71V2r9/fz6d8KOPPtqzZ8/ly5dv/c3MyspycXEB8P7774v7A3Vr165dW716dUhIiG4kMDIyMjg4mN9m2bFjh2iV/PrrrwBGjRolWoukPdqLSVSDBw/OyMhITk6eOnWq7mBAQEBBQYGNjY2vr2/gDYMGDRo8eLC/v7+VVafWhE1JSZk2bZpMJtu7d+9jtJK5uNLT0zdu3PjFF1+o1WoXFxdXV9fk5OQhIi6EXFJS4u3t3bt374qKCtEaJTdJnePdSFpaGgBfX98O40inT58uKCjQ/3qc3yDz8PC4cuWKnqciXVBWVjZo0CAAX331lfitOzs7A6isrBS/aULrjYqHL1O/aNGiDntPDh061M/PT/89ON95553HH3/86tWrM2fObG5u1vNs5F55eHiMHz8eQElJifitBwYGwuw3QTFWFKMiqaio+Pbbb62srBYsWCBQExYWFl9++WW/fv2OHj26bNkygVohd9FhJ5KUlJTly5cfP35chKa7xV5SxopiVCRxcXHNzc1RUVH80SOB9OrVa9euXfb29uvXr9+6datwDZHb6pBle/bsWb169eHDh8VvmoiJYlQMjLHNmzdDlD2Qhw0btnbtWgAvvvjiyZMnhW6OtNchy8TcJs/8t+QzYhSjYvjpp5+ys7P79esXGRkpQnMKhWLBggVNTU1PPvnktWvXRGiRcIGBgZaWlvn5+S0tLRC3hyjVXlIEFKPi0A0u3XY6vRA+/vjjESNG5OXlPffcc1qtVpxGiY2NjZ+fX1tbW15eHsSNUbqolxDFqOBKSkpSUlKsra11a1mKwM7OLikpyc3NLSUlJTY2VrR2Sfs4k8vl/Im11tZWodv19fW1t7cvKyurq6sTui3SAcWo4DZt2tTa2jpjxgy+/Y5o/P39d+zYYWlpuXLlyr1794rZdHfWPkbt7Ox8fHw0Gk1BQYHQ7VpYWPTr1w90e1QKFKPC0mq1/HlnEQaXbhUREfHWW29ptdr58+er1WrxC+iGOgz1iHnLkq7rpUIxKqzvv/8+Pz9fLpfz3XTF98Ybb3h7ew8aNKiyslKSArqbDlnGX4rTQ6QYlUqnntcmXcYHl1588UX9H1LqpMzMzMzMzKioKP4w/qVLl0pKSurq6nr06CFOAd0czXnqhqg3KqArV6788MMPtra2zz77rGiNrl27dsaMGW+//TZ/uWHDBsbY3Llz+TPXRGhyuVwmk+Xm5mo0GtBgffdAMSqgjRs3trW1RUdH32nLEIOrr6/fsWMHgKeffhqAWq3evn07AN3G5URo9vb23t7era2tfFiJYrQ7oBgVikajEX9wafv27bW1tePGjeNLDcXHx1dVVY0ePZqv60zE0f5+aFBQkEwmy8nJEWL2LmOs/Ro0AQEBNjY2RUVFNJwoMopRoezevbu4uHjgwIFhYWGiNbpp0ya0C25+Z1aSSQLdWfteoZOTU58+fZqbmwsLCw3bSn19/Zw5c2bPns1uLBlsaWnZt29fxhjdHhUZxahQtFpt7969w8PDdYNLtbW1s2fPPnjwoEAtHj169OTJk25ubtOnTwdw9uzZo0ePuri4REdHC9Qiua0OQz1CXGtnZmaOHDkyISFh//79WVlZuuO2trYAYmNji4uLDdgcuTuKUaHY2tpeu3YtISFB9798w4YNCQkJs2bNKi0tFaJF3vdcsGAB/13asGEDgOeee063ZRARx23nPBkwRpOTkx966KGLFy8OGDDgyJEj/fv3x43O6blz53r16rVjxw4/P7/HHnssMTGxra3NUO2SO5J00WhzptVqeTfwvvvuq6+vZ4xpNBq+vcfo0aObm5sN21x1dbWDg4NMJrt8+TJjTDfDiTZeFh9fWGvw4MH85datW//yl7+kpqbqf2aNRhMTE8Ovb5566in+/4oxdvnyZb5nibOz88qVK2fNmqXbQDQwMPDdd98tLS3Vv3VyJxSjAqqrq+NDPU899RQ/UlZW5uvrC+Dvf/+7Ydv6z3/+AyAyMpK//PTTTwGMGTPGsK2QzuBPtdvZ2bW1tRnwtFevXo2IiABgZWWlVCp1x5OTk/mGhiEhIbpvzdLS0nfffZcviQ/AxsZm1qxZh37+mdHesQKgGBVWZmYm7xV+9NFH/MiRI0d4T6H9PuP6452RpKQk/vL++++HRJsCEcZYnz59ABQUFBjqhMePHw8ICADg4eGRlpbGD2o0mpUrV/Jlw6ZNm1ZdXd3hU21tbampqdHR0dbW1gBWjR3LgoOZUsnKyw1VGGEUoyL45ptvZDKZtbX1r7/+yo+sW7cOgJOTk6GuuPn+up6ennzv5WPHjgFwc3NTq9UGOT+5V3x6xr59+wxyNpVKxb96H3nkkeLiYn6wvLyc3yPindO7b4lYWFi4atWqouHDGcAAZmfH/vIXdvCgQcojFKNiePXVV3nMFRUV8SPz5s0D0L9//1t7EF3wzDPPAHjzzTf5S77d0z/+8Q/9z0y65rXXXpPL5W5ubjExMXl5eV0+j1qtfv755/mFuUKh4F+TjLETJ07wzqm7u/s9hHVbG0tNZdHRzMrqep6GhDClklVUXH9Dbi7bvJlt2cLy87tcczdEMSqG1tbWMWPGAHj44Yf5b0JjYyOfEj9t2jQ9t1auqKiws7OzsLDgv67V1dWOjo4ymezSpUuGqZ50Ce8q8t7i9OnTf/jhh3u9VZqVlRUaGsovXOLj43XHdZ3TsLAw3RfzvcnLYytWMC+v62Hq4MBWr2b79rEHHmDbtrHPPmPDhrH9+7ty5m6JYlQkpaWlfDO7ZcuW8SNZWVl8ZOC9997T58wffvghgKioKP6S3zGIiIjQt2KitxMnTigUCnt7e56nPj4+MTExV65c6cxn9+zZ07NnTz5wdOHCBX5QrVbPnz9f1znVd76HRsNSU1lUFJPJWHw8e+ABlpFx/a9OnmQPPaTXybsTilHx6AaXdu7cyY989913FhYWlpaWP/74Y9fOqdVq+bTB7777jh+57777ACQmJhqmaKK30tJSpVLZftA8Ojo6NTX17lchhw8ftrGxeeKJJ6qqqviR/Pz8ESNGALC3t//ss88MWWJWFlOrmYfHHw66udGwfidRjIqKT0tqP7i0YsUKPvzayU5KB0VFRSEhIX5+fhqNhjHGH5HSjTUR49Fh0BxAcHCwUqm8evXqnT6Snp6ui9qUlBRXV1f+qbNnzwpSorv7zdxsa+uYquTOKEbF1mFwqa2t7fHHHwcwatSopqamLpxQq9XqJtbMnTsXwIoVKwxZMTGokpISpVLZt29fHqa2trZ375xqtVqlUslnNU2dOlXXOTW8yMib90N//JFNmSJUQ2aHYlRstw4ulZeX+/v7A1iyZIk+Z66qqnJwcOB7qBmmViIYXeeUr67N74EqlcoK3aA5Y4yxioqKiRMnAuB7ahl2Pn9H586xYcNYTAx77TU2fDi7eFHAtswLxagEdINLH3zwAT9y6tQpuVx+6NAhfU67Zs0aAJMnTzZEjUQkRUVFSqWSf4/yZ59455QxdvLkSb5LnZub208//SRGNQ0N7PBhlpLCMjPFaM5cUIxKIzk5WSaTWVpa6n499L+bOXDgQAC7d+/WuzoitpaWlqSkpMjISH7xzp+F50vMjBo1yoBPQ/25bdsYwObMEa9F0ydjNxYrJCJbsWJFbGysh4dHeno6f9C+86qqqn6/4cKFCxkZGVlZWTKZTCaTlZeX664TickpLCyMi4vbsGFDQ0ODm5vbxIkT161bp1tnRAyHDiE8HA8+iGPHxGvUxFGMSkar1U6ePPnHH3986KGHfvnll9v+qrS1tV25ciUnJyf7Bv7n2y5v7ujo2NDQ8Omnn9I6zaaupaXlzJkzgYGBvXv3FrvtsjJ4esLFBVVVYjdtsihGpVRZWTlixIjc3NyXXnrp3//+d0FBwe9/dPHixcbGxls/6OrqGniLo0ePPv3009bW1vv37xdzyX1ibnr2RG0tKiogfoibJopRiR07duzRRx/lu0jedoVdb2/voKAguVweFBSk+wN/vuVWS5Ys+fjjj/38/NLT00XbR4+Ym/vvx6lT+O03jBoldSmmgWJUeuHh4efOnautreV9zEGDBg0ePJh3MPv3739PGyO3trZOmDDhwIED48eP/+mnnywtLYUrm5itWbOQmIjt2/HMM1KXYhpoLEJi165dS09Pb2lpuXTpEn+sUx/W1tYJCQn3339/Wlram2++GRsba5AiSfcSFAQAtFFzp9FeTBLbunVrU1PTpEmT9M9QztPT8+uvv7a2tlYqlUlJSQY5J+le5HIAoIfFukQAAARySURBVO1FO41iVEqMsY0bN8LQeyA//PDDsbGxjLFPV6zQtts2kpBO4TFKvdFOo3ujUkpLS5swYYKvr29eXp5h72MyxjYvWTI/Pt7Cywu//QZHRwOenJi35uLiX5955qKV1dLUVKlrMQ3UG5US3xJ50aJFBh8LkslkC5VKiz59cP48Fi407MmJebPx8pp+7NjL+/bV1NRIXYtpoBiVTHl5+e7du62srPieH4bn5IRdu9CjB+Lj8dFHgjRBzJFMJuOro+bQ7dHOoRiVTFxcXHNz89SpU/mq+IIICcG2bZDJsHw5DhwQqhVidoKCggBk0+3RzqEYlYhWq/75Zxh6cOk2pk/Hq6+itRWzZqG4WNi2iLmgGL0nFKMSSU1956efaseN0218JiClEmPGoLQU0dFobRW8OWL65HI56KK+0yhGJfLppwCc2y2MJiArK8THw9sbhw/jgw8Eb46YPuqN3hOa8CSFkhIEBEAmQ0EB+vQRqdFDh/Dvf2PzZvTsCf7w/r1MD6itre2wylRubm5OTo5uZyFiTvLy8vr16+ft7V1UVCR1LSaAHgaVwqZNaG3F7NniZSiAsDCEheHaNcycicpKAHB1hUoFN7eO77x2DdnZyMlBdvb/q6zcc+xYdnZ2eXn5rafMzc011MNXxKj4+/vb2tqWlJQ0NDQ40qTjP0MxKrq2NmzZAgCSrAq6fDkiIvDCCwDwySdYvhxvvokDB3hoXk/PdgtNXgoLO3LkCAAbGxtfX1/donx8/ZSAgAAJfgQiPAsLi379+lVWVpaWlvL7pOQu6KJedN99hyeegFyOrCzIZGK37uWFK1fAl8fXaODnhzfewEsv/eE9zs4ICoJcjqCgU6GhNV5eQUFBPj4+MvGrJdJpbm7mu5iQP0W9UdGpVADw4osSZChj0Gig22LEygoaDYYPx1NPXc/N4GDI5e1vNQwXu0RiLChDO496o+K6cgX9+sHKCleuQJJllQcMQFoavL0BoLgY48cjM1OCMggxIzThSVwFBQgKwsyZ0mQogOXLsXgxcnKQk4PFi7F8uTRlEGJGqDcqlj178Msv6NED//M/8PFBjx6SVbJ7N3bvBoDp0/HEE5KVQYi5oBgVxb/+hTNnsGwZKirw+uuIi8PIkVLXRAgxDIpR4bW0IDAQWVmwtweAH37Apk2gdekJMRc0Ui+AkpKbczCzs7FkCXx8rmcogGHDaFSHEHNCMdp1Wq22sLAwOzs7oKhIfv78zdxsaPjD+558EvX1N1/W1eEO2yMTQkwRxWhnFRcXZ2Rk/N5OZmZmQ0MDgP8bO/aNn3+++dbevXXT1xEUhPBw2Nri1CkMHw4AX32FyEhpfgZCiADo3minpKSkTJ069dbjffr0CQ4OnhcWttDJ6XpoyuVwde34vtOnsXAhBg5EZSVsbfH553ByEqNuQojwKEY75dKlSxEREUE3yOVy/gene0rD/Hz06HGbkCWEmDKKUUII0Qs9xUQIIXqhGCWEEL1QjBJCiF4oRgkhRC8Uo4QQopf/DxoIbuxd3cxgAAACQnpUWHRyZGtpdFBLTCByZGtpdCAyMDIwLjA5LjEAAHice79v7T0GIOABYkYGCFAEYiUgbmBkY0gAiTNzMGgAaWYmNgcwzcLmkAGimRmRGFAZsA4mJhjNzqAAEocZhC6NoEHKgLpBpjCxsENoZsL2wV0GswCnOEwjWICZEdVFyApwmMANDB9GJgUm5gwmZpYEFtYMJlY2BjZ2BnYOBg5OBg4uBi5uBm4eBh5eBl4+Bj5+DmZeAQYBwQwmQaEEIeEMJmERBRFRBmGxBDFxBXEJBjHJDCZeKQYpaQYpGQZW2QRZuQwmOXkFeQUGOaYESYEEESY2JjlZVhZmNkEhYTFJAXFY5DAo6nbwOJpGxxwAcZ4ZcTrumfRrP4j95ulTh8kzNoHZcwQOOPAkioDV7Jkw2eHD8Qdgcb1nPQ7sE7v2gdjzzjg4zA6fagdiN5UbO0ytfGwPYne832J/f6eoA4jt7zrLftvpEjBb7lKOnYXXVrAay9Qp+xNd/oLZdndFD9yWagez/9zJO8Aieh7MfnHf64BDlwNY75VHagf+H+kBszlyOw/YrMkA2/vl644D6y/PsgGx676cOqA3ZT3YnbJZaw/sLFEGu395xv4D5SU1YPakj+8OnHrSAmYnHao+MEdeDMxesNbhgN2pRDB7ovnh/UqlIWC239HwA5X1JWAzY98tPjCtnRvshtczVh3InZkFZq87fPLA8qcLwG6u373JoZ1LHcyOs77vsGHjDlsQO6KDyXHBlkawuGLaPwf1N/JgvWIAClCmGs682PgAAAImelRYdE1PTCByZGtpdCAyMDIwLjA5LjEAAHicnVZJjtswELz7FfyAiV5J9jme0yATIIf8Iff8HymaEq1DAsy0IRjVaqtQvbCsW5mfn4/333/K/sjjdiuqRa0U+ucVEeWXENFt/n7UPlQA7lrNlOc9qshS+Vb+R3G9DhbvFpOFa3hvF5Yfn2fplYfpYtFwzWlpeEJoskhVEOZYrDZmW1q6t2RfrBo9e3qnKp1ajkUrsfh6VoJ7jkXqIJuTLqio92RFXI21LT7FkC4sH19hEXJZtQ15bk5CC7qhrKeqltsXzJc9DpboIzejuWuQsFjI3XMsWrvKMaPGfWRZuEUcm0PJE4CKGq2TbDiNQalJ3w0ba0dPRbrktDhcik4FPHL7cm+YkZz+0pPnCFqMQ1aH3I2yLDHY17TCLOeY916tdVp9ZpdkX6BAtZ9b7Mm+TJd6OhyQy6BcRYxnnReLciRPo8752vLdcM95HWbEImN5HQ9K764qH17Xmicrwu7SqgPe6T23dcWrDz/8oLnkXAr/sNHbqSCSfcE7A/s4fJdYr+79pXeG8OUMUi1GyjHRVnl+zwBIdwBkz1szAPKdAWo7A9R3BmjsACh2EIXpGvAZAPFWAMRbAdDc4yOwwlsB0Py/0Z3ZCoB4nNqAeCsAmm9GKwMkWwGQbAUzsxUAyVYAJH7N7B5AgWwFQDKuwasHrejuAZDySQCkWwGQbgXzNy857VXp91LePh63v9WBtA0KNlBvAAABgHpUWHRTTUlMRVMgcmRraXQgMjAyMC4wOS4xAAB4nCVSOW7EMAz8SoA0NiALPEUSLt1nH+Fv5PEZKlsY2OExB/V8Xn7f93ien+M5H/yO5/vnPF5p9POc/x85ny6f78uf5+v3yBmp4uPSaaYU487pYaXj4lkea9wxOU2lAS2XHPeaREJokamoybhtLmbbPeGLGzBSHhdNCVqKIZ3EYjWAFK+0cctMMo6BmQj2cfM0Vl8Da7UsGhBysaEzpTXdGFaWHjFeCgCE7KUGpCIXlrSoLgIhd4cWeAsV6Z7FsLsRXlWCxURptqcWhdcwmCzqPQahBjXglBCYvBxhwTkQ4swquLoW2ER9pxMcvNuMsRsr3W3/r2SXzrjMOp0rpi0IGCBBpf4llWrYDhW0uSFiWj3nEhXbrYuzNqRc2642b0dPuJf3BUHIIunIkZM6WCCqjKh1rrXv07rJuTMRDwLi09O177PAQdZnrmj7VLAr+2mwp1Qny4oXgadRVhAm0yqjxvn7B9P5g1W8rrS/AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<rdkit.Chem.rdchem.Mol at 0x154b8ea40bc0>"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mol = Chem.MolFromSmiles(smi_train[-1])\n",
    "mol"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ff156b2d",
   "metadata": {},
   "source": [
    "# 3. Investiagate how can we featurize molecular structure"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f2ccefbf",
   "metadata": {},
   "source": [
    "## 1) Atom (node) features \n",
    "* Atom symbol -- ex) carbon: C, oxygen: O\n",
    "* Atom degree\n",
    "* Number of hydrogens attatched to a given atom (implicit hydrogen model)\n",
    "* Implicit valence \n",
    "* Whether a given atom has an aromatic character or not"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "3772986b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "33"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "atom_list = mol.GetAtoms()\n",
    "num_atoms = len(atom_list)\n",
    "num_atoms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "b7709ec8",
   "metadata": {},
   "outputs": [],
   "source": [
    "atom_ = atom_list[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "294b1468",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<rdkit.Chem.rdchem.Atom at 0x154b8ea06300>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "atom_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "9bb31c3c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'C'"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "atom_.GetSymbol()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "de0d21ae",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "atom_.GetDegree()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "86f5c39f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "atom_.GetTotalNumHs()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "0f5652bc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "atom_.GetImplicitValence()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "a9a71fef",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "False"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "atom_.GetIsAromatic()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e6dbf3e2",
   "metadata": {},
   "source": [
    "## 1) Bond (edge) features \n",
    "* Bond types -- Single, Double, Triple, Aromatic\n",
    "* Whether a given bond is conjugated or not\n",
    "* Whether a given bond has a ring character"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "b76d2104",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "34"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bond_list = mol.GetBonds()\n",
    "num_bonds = len(bond_list)\n",
    "num_bonds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "c87d7f7d",
   "metadata": {},
   "outputs": [],
   "source": [
    "bond_ = bond_list[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "c1275559",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "rdkit.Chem.rdchem.BondType.SINGLE"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bond_.GetBondType()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "48e45968",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "False"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bond_.GetIsConjugated()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "5715384b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "False"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bond_.IsInRing()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e0a44354",
   "metadata": {},
   "source": [
    "# 4. Put them altogether -- define a series of functions that featurize molecular graph"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "59f60eed",
   "metadata": {},
   "outputs": [],
   "source": [
    "# GLOBAL VARIABLE\n",
    "ATOM_VOCAB = [\n",
    "    'C', 'N', 'O', 'S', 'F',\n",
    "    'H', 'Si', 'P', 'Cl', 'Br',\n",
    "    'Li', 'Na', 'K', 'Mg', 'Ca',\n",
    "    'Fe', 'As', 'Al', 'I', 'B',\n",
    "    'V', 'Tl', 'Sb', 'Sn', 'Ag',\n",
    "    'Pd', 'Co', 'Se', 'Ti', 'Zn',\n",
    "    'Ge', 'Cu', 'Au', 'Ni', 'Cd',\n",
    "    'Mn', 'Cr', 'Pt', 'Hg', 'Pb',\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "5e3b92d2",
   "metadata": {},
   "outputs": [],
   "source": [
    "def one_of_k_encoding(x, vocab):\n",
    "    if x not in vocab:\n",
    "        x = vocab[-1]\n",
    "    return list(map(lambda s: float(x==s), vocab))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "c8ba797f",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_atom_feature(atom):\n",
    "    atom_feature = one_of_k_encoding(atom.GetSymbol(), ATOM_VOCAB)\n",
    "    atom_feature += one_of_k_encoding(atom.GetDegree(), [0, 1, 2, 3, 4, 5])\n",
    "    atom_feature += one_of_k_encoding(atom.GetTotalNumHs(), [0, 1, 2, 3, 4])\n",
    "    atom_feature += one_of_k_encoding(atom.GetImplicitValence(), [0, 1, 2, 3, 4, 5])\n",
    "    atom_feature += [atom.GetIsAromatic()]\n",
    "    return atom_feature"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "8034abe3",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_bond_feature(bond):\n",
    "    bt = bond.GetBondType()\n",
    "    bond_feature = [\n",
    "        bt == Chem.rdchem.BondType.SINGLE,\n",
    "        bt == Chem.rdchem.BondType.DOUBLE,\n",
    "        bt == Chem.rdchem.BondType.TRIPLE,\n",
    "        bt == Chem.rdchem.BondType.AROMATIC,\n",
    "        bond.GetIsConjugated(),\n",
    "        bond.IsInRing()\n",
    "    ]\n",
    "    return bond_feature"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "2fafa31c",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_molecular_graph(smi):\n",
    "    mol = Chem.MolFromSmiles(smi)\n",
    "    graph = dgl.DGLGraph()\n",
    "    \n",
    "    atom_list = mol.GetAtoms()\n",
    "    num_atoms = len(atom_list)\n",
    "    graph.add_nodes(num_atoms)\n",
    "    \n",
    "    atom_feature_list = [get_atom_feature(atom) for atom in atom_list]\n",
    "    atom_feature_list = torch.tensor(atom_feature_list, dtype=torch.float64)\n",
    "    graph.ndata['h'] = atom_feature_list\n",
    " \n",
    "    bond_list = mol.GetBonds()\n",
    "    bond_feature_list = []\n",
    "    for bond in bond_list:\n",
    "        bond_feature = get_bond_feature(bond)\n",
    "        \n",
    "        src = bond.GetBeginAtom().GetIdx()\n",
    "        dst = bond.GetEndAtom().GetIdx()\n",
    " \n",
    "        # DGL graph is undirectional\n",
    "        # Thus, we have to add edge pair of both (i,j) and (j, i)\n",
    "        # i --> j\n",
    "        graph.add_edges(src, dst)\n",
    "        bond_feature_list.append(bond_feature)\n",
    "\n",
    "        # j --> i\n",
    "        graph.add_edges(dst, src)\n",
    "        bond_feature_list.append(bond_feature)\n",
    "\n",
    "    bond_feature_list = torch.tensor(bond_feature_list, dtype=torch.float64)\n",
    "    graph.edata['e_ij'] = bond_feature_list\n",
    "    return graph"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "a3c2b299",
   "metadata": {},
   "outputs": [],
   "source": [
    "smi_ = smi_train[-1]\n",
    "g_ = get_molecular_graph(smi_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "e2143491",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Graph(num_nodes=33, num_edges=68,\n",
       "      ndata_schemes={'h': Scheme(shape=(58,), dtype=torch.float64)}\n",
       "      edata_schemes={'e_ij': Scheme(shape=(6,), dtype=torch.float64)})"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "ec706acc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[1., 0., 0.,  ..., 0., 0., 0.],\n",
       "        [0., 0., 1.,  ..., 0., 0., 0.],\n",
       "        [1., 0., 0.,  ..., 0., 0., 1.],\n",
       "        ...,\n",
       "        [1., 0., 0.,  ..., 0., 0., 1.],\n",
       "        [0., 0., 1.,  ..., 0., 0., 0.],\n",
       "        [1., 0., 0.,  ..., 0., 0., 0.]], dtype=torch.float64)"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g_.ndata['h']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "1302cd8f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([33, 58])"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g_.ndata['h'].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "d2f62537",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 1., 0.],\n",
       "        [1., 0., 0., 0., 1., 0.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 1., 0., 0., 0.],\n",
       "        [0., 0., 1., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [1., 0., 0., 0., 1., 0.],\n",
       "        [1., 0., 0., 0., 1., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [1., 0., 0., 0., 1., 0.],\n",
       "        [1., 0., 0., 0., 1., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [1., 0., 0., 0., 1., 0.],\n",
       "        [1., 0., 0., 0., 1., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [1., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.],\n",
       "        [0., 0., 0., 1., 1., 1.]], dtype=torch.float64)"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g_.edata['e_ij']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "d39b4063",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([68, 6])"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g_.edata['e_ij'].shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f35ee2fe",
   "metadata": {},
   "source": [
    "# GNN Practice"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ff34b94b",
   "metadata": {},
   "source": [
    "# 1. GCN, GIN and GIN+E"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "625e8b07",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.nn as nn \n",
    "import torch.nn.functional as F\n",
    "\n",
    "import dgl.function as fn\n",
    "from dgl.nn.functional import edge_softmax"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "afd62c56",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-0.0074, -0.2408,  0.2086,  ..., -0.2388, -0.0149, -0.0177],\n",
       "        [-0.1628,  0.0205,  0.0198,  ..., -0.0343,  0.0858, -0.3845],\n",
       "        [-0.0555, -0.2638,  0.2622,  ...,  0.1037,  0.1383, -0.0746],\n",
       "        ...,\n",
       "        [-0.0555, -0.2638,  0.2622,  ...,  0.1037,  0.1383, -0.0746],\n",
       "        [-0.1628,  0.0205,  0.0198,  ..., -0.0343,  0.0858, -0.3845],\n",
       "        [-0.0074, -0.2408,  0.2086,  ..., -0.2388, -0.0149, -0.0177]],\n",
       "       grad_fn=<MmBackward>)"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "h0 = g_.ndata['h'].float()\n",
    "embedding_node = nn.Linear(58, 64, bias=False)\n",
    "embedding_edge = nn.Linear(6, 64, bias=False)\n",
    "h0 = embedding_node(h0)\n",
    "g_.ndata['h'] = h0\n",
    "h0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "95163d9f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([33, 64])"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "h0.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "c0bd9761",
   "metadata": {},
   "outputs": [],
   "source": [
    "e_ij = embedding_edge(g_.edata['e_ij'].float())\n",
    "g_.edata['e_ij'] = e_ij"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "48f73b8e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([68, 64])"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "e_ij.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "86477099",
   "metadata": {},
   "source": [
    "## 1) Graph Convolution"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8007d018",
   "metadata": {},
   "source": [
    "$ h_{i}^{(l+1)} = \\sigma (W^{(l+1)} \\sum_{j \\in N(i)} h_{j}^{(l)})$"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "78d45dd2",
   "metadata": {},
   "source": [
    "### u_ : sum aggregation of the adjacent node features "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "9a6e046f",
   "metadata": {},
   "outputs": [],
   "source": [
    "g_.update_all(fn.copy_u('h', 'm'), fn.sum('m', 'u_'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "d3bbb032",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-0.1628,  0.0205,  0.0198,  ..., -0.0343,  0.0858, -0.3845],\n",
       "        [-0.0629, -0.5046,  0.4708,  ..., -0.1351,  0.1234, -0.0923],\n",
       "        [-0.3833, -0.4679,  0.4130,  ..., -0.1529,  0.1644, -0.7081],\n",
       "        ...,\n",
       "        [-0.3833, -0.4679,  0.4130,  ..., -0.1529,  0.1644, -0.7081],\n",
       "        [-0.0629, -0.5046,  0.4708,  ..., -0.1351,  0.1234, -0.0923],\n",
       "        [-0.1628,  0.0205,  0.0198,  ..., -0.0343,  0.0858, -0.3845]],\n",
       "       grad_fn=<GSpMMBackward>)"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "u_ = g_.ndata['u_']\n",
    "u_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "564308e7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([33, 64])"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "u_.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "80cadb15",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([33, 64])"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g_.ndata['u_'].shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1b10d053",
   "metadata": {},
   "source": [
    "### Applying linear layer, non-linear activation, layer normalization, and dropout"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "dac59918",
   "metadata": {},
   "outputs": [],
   "source": [
    "linear_ = nn.Linear(64, 64, bias=False)\n",
    "norm_ = nn.LayerNorm(64)\n",
    "act_ = F.relu\n",
    "prob_ = 0.2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "fd661175",
   "metadata": {},
   "outputs": [],
   "source": [
    "h_ = linear_(u_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "e84970df",
   "metadata": {},
   "outputs": [],
   "source": [
    "h_ = act_(h_) + h0 # non-linear activation (ReLU) and skip-connection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "7e3514ea",
   "metadata": {},
   "outputs": [],
   "source": [
    "h_ = norm_(h_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "753687ac",
   "metadata": {},
   "outputs": [],
   "source": [
    "h_ = F.dropout(h_, p=prob_, training=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "d97215d7",
   "metadata": {},
   "outputs": [],
   "source": [
    "# g_.ndata['h'] = h_ <-- update node feature, \n",
    "# but it will be omitted in this practice because we have to keep practice other GNN layers"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "78d695c7",
   "metadata": {},
   "source": [
    "## 2) Graph Isomorphism"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "31e0abfa",
   "metadata": {},
   "source": [
    "$ h_{i}^{(l+1)} = \\sigma (\\text{MLP}^{(l+1)} (\\sum_{j \\in N(i)} h_{j}^{(l)})) $"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "fe9fdf04",
   "metadata": {},
   "outputs": [],
   "source": [
    "class MLP(nn.Module):\n",
    "    def __init__(\n",
    "        self,\n",
    "        input_dim,\n",
    "        hidden_dim,\n",
    "        output_dim,\n",
    "        bias=True,\n",
    "        act=F.relu,\n",
    "    ):\n",
    "        super().__init__()\n",
    "        \n",
    "        self.input_dim = input_dim\n",
    "        self.hidden_dim = hidden_dim\n",
    "        self.output_dim = output_dim\n",
    "\n",
    "        self.act = act\n",
    "\n",
    "        self.linear1 = nn.Linear(input_dim, hidden_dim, bias=bias)\n",
    "        self.linear2 = nn.Linear(hidden_dim, output_dim, bias=bias)\n",
    "\n",
    "    def forward(self, h):\n",
    "        h = self.linear1(h)\n",
    "        h = self.act(h)\n",
    "        h = self.linear2(h)\n",
    "        return h"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3c449fb5",
   "metadata": {},
   "source": [
    "### u_ : sum aggregation of the adjacent node features, same as GCN practice"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "6d86a208",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([33, 64])"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g_.update_all(fn.copy_u('h', 'm'), fn.sum('m', 'u_'))\n",
    "u_ = g_.ndata['u_']\n",
    "u_.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "6ab7afb5",
   "metadata": {},
   "outputs": [],
   "source": [
    "### Applying linear layer, non-linear activation, layer normalization, and dropout"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "c4634c79",
   "metadata": {},
   "outputs": [],
   "source": [
    "mlp_ = MLP(\n",
    "    input_dim=64, \n",
    "    hidden_dim=64*4, \n",
    "    output_dim=64,\n",
    ")\n",
    "norm_ = nn.LayerNorm(64)\n",
    "act_ = F.relu\n",
    "prob_ = 0.2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "3a1e3cdb",
   "metadata": {},
   "outputs": [],
   "source": [
    "h_ = mlp_(u_)\n",
    "h_ = act_(h_) + h0 # non-linear activation (ReLU) and skip-connection\n",
    "h_ = norm_(h_)\n",
    "h_ = F.dropout(h_, p=prob_, training=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "781acb1c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# g_.ndata['h'] = h_ <-- update node feature, \n",
    "# but it will be omitted in this practice because we have to keep practice other GNN layers"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6b4d320b",
   "metadata": {},
   "source": [
    "## 3) Graph Isomorphism w/ edge features"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6047bba3",
   "metadata": {},
   "source": [
    "$ h_{i}^{(l+1)} = \\sigma (\\text{MLP}^{(l+1)} (\\sum_{j \\in N(i)} h_{j}^{(l)} + e_{ij}^{(l)})) $"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6891e5c3",
   "metadata": {},
   "source": [
    "### u_ : sum aggregation of the adjacent node features, same as GCN practice\n",
    "### e_ : sum aggregation of the edges connected to the center atoms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "id": "e44e60f5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([33, 64])"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g_.update_all(fn.copy_u('h', 'm'), fn.sum('m', 'u_'))\n",
    "u_ = g_.ndata['u_']\n",
    "u_.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "d2f2d488",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([33, 64])"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g_.update_all(fn.copy_edge('e_ij', 'm_e'), fn.sum('m_e', 'e_'))\n",
    "e_ = g_.ndata['e_']\n",
    "e_.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "a11e5f67",
   "metadata": {},
   "outputs": [],
   "source": [
    "h_ = g_.ndata['u_'] + g_.ndata['e_']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "2c35492a",
   "metadata": {},
   "outputs": [],
   "source": [
    "h_ = mlp_(u_)\n",
    "h_ = act_(h_) + h0 # non-linear activation (ReLU) and skip-connection\n",
    "h_ = norm_(h_)\n",
    "h_ = F.dropout(h_, p=prob_, training=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "fb1bf6c9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([33, 64])"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "h_.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5441236b",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}