docs/tutorials/pyg_integration.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "import torch\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import torch.nn.functional as F\n",
    "from tqdm.auto import tqdm"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## PyG integration\n",
    "\n",
    "<div class=\"admonition tip highlight\">\n",
    "    <p class=\"admonition-title\">Community contribution</p>\n",
    "    <p>Curious how one would run this tutorial on <a href=\"https://www.graphcore.ai/products/ipu\">Graphcore IPUs</a>? See this tutorial contributed by <a href=\"https://github.com/s-maddrellmander\">@s-maddrellmander</a>:\n",
    "    <a href=\"https://ipu.dev/lDvDHL\"><img alt=\"Run on Gradient\" src=\"https://camo.githubusercontent.com/c9931a1689c37ab786edd3e1e5f59b9a6f7d097628c4689ce2432563ef884524/68747470733a2f2f6173736574732e706170657273706163652e696f2f696d672f6772616469656e742d62616467652e737667\"></a></p>\n",
    "</div>\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As seen in the [molfeat integration tutorial](./integration.html), molfeat integrates easily with the PyTorch ecosystem. In this tutorial, we will demonstrate how you can integrate molfeat with [PyG](https://pytorch-geometric.readthedocs.io/en/latest/) for training SOTA GNNs.\n",
    "\n",
    "To run this tutorial, you will need to install `pytorch-geometric`.\n",
    "\n",
    "`mamba install -c conda-forge pytorch_geometric`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from molfeat.trans.graph.adj import PYGGraphTransformer\n",
    "from molfeat.calc.atom import AtomCalculator\n",
    "from molfeat.calc.bond import EdgeMatCalculator"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Featurizer\n",
    "\n",
    "We first start by defining our featurizer. We will use the `PYGGraphTransformer` from molfeat with atom and bond featurizers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "featurizer = PYGGraphTransformer(\n",
    "    atom_featurizer=AtomCalculator(), \n",
    "    bond_featurizer=EdgeMatCalculator()\n",
    ")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Dataset\n",
    "\n",
    "For the dataset, we will use the `Lipophilicity` dataset (LogD) from MoleculeNet, which contains experimental results of octanol/water distribution coefficient at pH=7.4\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_csv(\"https://deepchemdata.s3-us-west-1.amazonaws.com/datasets/Lipophilicity.csv\")"
   ]
  },
  {
   "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>CMPD_CHEMBLID</th>\n",
       "      <th>exp</th>\n",
       "      <th>smiles</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>CHEMBL596271</td>\n",
       "      <td>3.54</td>\n",
       "      <td>Cn1c(CN2CCN(CC2)c3ccc(Cl)cc3)nc4ccccc14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>CHEMBL1951080</td>\n",
       "      <td>-1.18</td>\n",
       "      <td>COc1cc(OC)c(cc1NC(=O)CSCC(=O)O)S(=O)(=O)N2C(C)...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>CHEMBL1771</td>\n",
       "      <td>3.69</td>\n",
       "      <td>COC(=O)[C@@H](N1CCc2sccc2C1)c3ccccc3Cl</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>CHEMBL234951</td>\n",
       "      <td>3.37</td>\n",
       "      <td>OC[C@H](O)CN1C(=O)C(Cc2ccccc12)NC(=O)c3cc4cc(C...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CHEMBL565079</td>\n",
       "      <td>3.10</td>\n",
       "      <td>Cc1cccc(C[C@H](NC(=O)c2cc(nn2C)C(C)(C)C)C(=O)N...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   CMPD_CHEMBLID   exp                                             smiles\n",
       "0   CHEMBL596271  3.54            Cn1c(CN2CCN(CC2)c3ccc(Cl)cc3)nc4ccccc14\n",
       "1  CHEMBL1951080 -1.18  COc1cc(OC)c(cc1NC(=O)CSCC(=O)O)S(=O)(=O)N2C(C)...\n",
       "2     CHEMBL1771  3.69             COC(=O)[C@@H](N1CCc2sccc2C1)c3ccccc3Cl\n",
       "3   CHEMBL234951  3.37  OC[C@H](O)CN1C(=O)C(Cc2ccccc12)NC(=O)c3cc4cc(C...\n",
       "4   CHEMBL565079  3.10  Cc1cccc(C[C@H](NC(=O)c2cc(nn2C)C(C)(C)C)C(=O)N..."
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Since training a network with PyTorch requires defining a dataset and dataloader, we can define our custom dataset that will take **(1)** the SMILES, **(2)** the LogD measurement, and **(3)** our molfeat transformer as input to generate the data point we need for model training."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import Dataset\n",
    "from torch.utils.data import DataLoader\n",
    "from torch_geometric.utils import degree\n",
    "\n",
    "\n",
    "class DTset(Dataset):\n",
    "    def __init__(self, smiles, y, featurizer):\n",
    "        super().__init__()\n",
    "        self.smiles = smiles\n",
    "        self.featurizer = featurizer\n",
    "        self.featurizer.auto_self_loop()\n",
    "        self.y = torch.tensor(y).unsqueeze(-1).float()\n",
    "        self.transformed_mols = self.featurizer(smiles)\n",
    "        self._degrees = None\n",
    "\n",
    "    @property\n",
    "    def num_atom_features(self):\n",
    "        return self.featurizer.atom_dim\n",
    "\n",
    "    @property\n",
    "    def num_output(self):\n",
    "        return self.y.shape[-1]\n",
    "    \n",
    "    def __len__(self):\n",
    "        return len(self.transformed_mols)\n",
    "    \n",
    "    @property\n",
    "    def num_bond_features(self):\n",
    "        return self.featurizer.bond_dim\n",
    "    \n",
    "\n",
    "    @property\n",
    "    def degree(self):\n",
    "        if self._degrees is  None:\n",
    "            max_degree = -1\n",
    "            for data in self.transformed_mols:\n",
    "                d = degree(data.edge_index[1], num_nodes=data.num_nodes, dtype=torch.long)\n",
    "                max_degree = max(max_degree, int(d.max()))\n",
    "            # Compute the in-degree histogram tensor\n",
    "            deg = torch.zeros(max_degree + 1, dtype=torch.long)\n",
    "            for data in self.transformed_mols:\n",
    "                d = degree(data.edge_index[1], num_nodes=data.num_nodes, dtype=torch.long)\n",
    "                deg += torch.bincount(d, minlength=deg.numel())\n",
    "            self._degrees = deg\n",
    "        return self._degrees\n",
    "\n",
    "    def collate_fn(self, **kwargs):\n",
    "        # luckily the molfeat featurizer provides a collate functoin for PyG\n",
    "        return self.featurizer.get_collate_fn(**kwargs)\n",
    "    \n",
    "    def __getitem__(self, index):\n",
    "        return self.transformed_mols[index], self.y[index]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = DTset(df.smiles.values, df.exp.values, featurizer)\n",
    "generator = torch.Generator().manual_seed(42)\n",
    "train_size = int(0.8 * len(dataset))\n",
    "test_size = len(dataset) - train_size\n",
    "train_dt, test_dt = torch.utils.data.random_split(dataset, [train_size, test_size], generator=generator)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "BATCH_SIZE = 64\n",
    "train_loader = DataLoader(train_dt, batch_size=BATCH_SIZE, shuffle=True, collate_fn=dataset.collate_fn(return_pair=False))\n",
    "test_loader = DataLoader(test_dt, batch_size=BATCH_SIZE, shuffle=False, collate_fn=dataset.collate_fn(return_pair=False))\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Network + Training\n",
    "We are almost ready to go, we just need to define our GNN. Here we use [PNA](https://arxiv.org/abs/2004.05718) as our GNN."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch_geometric.nn.models import PNA\n",
    "from torch_geometric.nn import global_add_pool\n",
    "\n",
    "DEVICE = \"cpu\"\n",
    "NUM_EPOCHS = 10\n",
    "LEARNING_RATE = 5e-4\n",
    "PNA_AGGREGATORS = ['mean', 'min', 'max', 'std']\n",
    "PNA_SCALERS = ['identity', 'amplification', 'attenuation']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = PNA(in_channels=dataset.num_atom_features, \n",
    "                    hidden_channels=128, \n",
    "                    num_layers=3,\n",
    "                    out_channels=dataset.num_output, \n",
    "                    dropout=0.1, \n",
    "                    act=\"relu\",\n",
    "                    edge_dim=dataset.num_bond_features,\n",
    "                    aggregators = PNA_AGGREGATORS,\n",
    "                    scalers = PNA_SCALERS,\n",
    "                    deg=dataset.degree,\n",
    ")\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=LEARNING_RATE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a0255ecabd0a40d6a8ec611478ca3ae6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/10 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Train\n",
    "model = model.to(DEVICE).float()\n",
    "model.train()\n",
    "with tqdm(range(NUM_EPOCHS)) as pbar:\n",
    "    for epoch in pbar:\n",
    "        losses = []\n",
    "        for data in train_loader:\n",
    "            data = data.to(DEVICE)\n",
    "            optimizer.zero_grad()\n",
    "            out = model(data.x, data.edge_index, edge_attr=data.edge_attr)\n",
    "            out = global_add_pool(out, data.batch)\n",
    "            loss = F.mse_loss(out.squeeze(), data.y)\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "            losses.append(loss.item())\n",
    "        pbar.set_description(f\"Epoch {epoch} - Loss {np.mean(losses):.3f}\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Testing\n",
    "\n",
    "We can now test our model. For the simplicity of this tutorial, no hyper-parameter search and evaluation of the best atom/bond featurization was performed. This inevitably impacts the performance. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import r2_score, mean_absolute_error\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "model.eval()\n",
    "test_y_hat = []\n",
    "test_y_true = []\n",
    "with torch.no_grad():\n",
    "    for data in test_loader:\n",
    "        data = data.to(DEVICE)\n",
    "        out = model(data.x, data.edge_index, edge_attr=data.edge_attr)\n",
    "        out = global_add_pool(out, data.batch)\n",
    "        test_y_hat.append(out.detach().cpu().squeeze())\n",
    "        test_y_true.append(data.y)\n",
    "\n",
    "test_y_hat = torch.cat(test_y_hat).numpy()\n",
    "test_y_true = torch.cat(test_y_true).numpy()\n",
    "\n",
    "r2 = r2_score(test_y_true, test_y_hat)\n",
    "mae = mean_absolute_error(test_y_true, test_y_hat)\n",
    "\n",
    "plt.scatter(test_y_true, test_y_hat)\n",
    "_ =plt.gca().annotate(\n",
    "    \"$R2 = {:.2f}$\\nMAE = {:.2f}\".format(r2, mae), \n",
    "    xy=(0.05,0.9),  \n",
    "    xycoords='axes fraction', \n",
    "    size=8,\n",
    "    bbox=dict(boxstyle=\"round\", fc=(1.0, 0.7, 0.7), ec=\"none\")\n",
    ")"
   ]
  }
 ],
 "metadata": {
  "interpreter": {
   "hash": "25cc46219716499d7b4b65986f7c346bc1ce4d9f3ea6745d62d4866f9ff86342"
  },
  "kernelspec": {
   "display_name": "Python 3.9.13",
   "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": "3.10.10"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}