HACNet.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU",
    "gpuClass": "standard"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/gregory-kyro/HAC-Net/blob/main/HACNet.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "![hacnet_logo-removebg-preview (2).png](data:image/png;base64,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)"
      ],
      "metadata": {
        "id": "UR-5lYfsw663"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**This Colab notebook allows you to easily and quickly predict the binding affinity between a protein and ligand using [HACNet v1.4.2](https://github.com/gregory-kyro/HAC-Net). For more details, check out the associated [journal article](https://pubs.acs.org/doi/10.1021/acs.jcim.3c00251) or [open-access article](https://arxiv.org/abs/2212.12440). Additionally, the software is available as open-source on [GitHub](https://github.com/gregory-kyro/HAC-Net).**\n",
        "\n",
        "**For a complete tutorial on how to use this notebook to predict $pK_{D}$, please navigate to this [demo video](https://www.linkedin.com/feed/update/urn:li:activity:7059553793595555840/).**\n",
        "\n",
        "**It is important to note that since HAC-Net was trained on docked complexes with nontrivial binding affinities, it should only be used to score protein-ligand pairs that bind.**\n",
        "\n",
        "**For details regarding the implementation in this notebook, please see [this document](https://github.com/gregory-kyro/HAC-Net/blob/main/HACNet/predict_pkd_info.txt).**\n",
        "\n",
        "**Feel free to reach out via email with any questions or concerns:**\n",
        "\n",
        "**gregory.kyro@yale.edu** &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; **rafi.brent@yale.edu**"
      ],
      "metadata": {
        "id": "VI5OVz1dw_cB"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**In order to properly run this notebook:**\n",
        "\n",
        "**Step 1 - run the first cell to begin setting up the environment of the notebook (this will crash the runtime)**\n",
        "\n",
        "**Step 2 - after the notebook has crashed, run the first cell again to finish setting up the notebook environment**\n",
        "\n",
        "**Step 3 - upload protein (pdb) and ligand (mol2) files to the notebook, as shown in the demo video**\n",
        "\n",
        "**Step 4 - type the names of the uploaded files into the second cell, select the type of output you would like to receive, and run the second cell**"
      ],
      "metadata": {
        "id": "1R6k5ceVxIZd"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": true,
        "id": "qBebTpX4MHCv"
      },
      "outputs": [],
      "source": [
        "#@title Set Up Environment { display-mode: \"form\" }\n",
        "\n",
        "#@markdown **Please execute this cell by pressing the *Play* button on the left\n",
        "#@markdown to set up the environment of the notebook to run HACNet. This will\n",
        "#@markdown initially crash the notebook to reconfigure the environment --\n",
        "#@markdown please then press *Play* again.**\n",
        "\n",
        "# hide outputs\n",
        "%%capture\n",
        "\n",
        "# install condacolab\n",
        "!pip install -q condacolab\n",
        "\n",
        "# import condacolab\n",
        "import condacolab\n",
        "\n",
        "# install condacolab environment\n",
        "condacolab.install()\n",
        "\n",
        "# import condacolab\n",
        "import condacolab\n",
        "\n",
        "# check that the environment has been initialized properly\n",
        "condacolab.check()\n",
        "\n",
        "# install openbabel\n",
        "!conda install -c conda-forge openbabel\n",
        "\n",
        "# install pymol\n",
        "!mamba install pymol-open-source --yes\n",
        "\n",
        "# install biopandas\n",
        "!pip install biopandas\n",
        "\n",
        "# import torch\n",
        "import torch\n",
        "\n",
        "# install necessary torch packages\n",
        "!pip install torch-geometric torch-scatter torch-sparse \\\n",
        " -f https://data.pyg.org/whl/torch-{torch.__version__}.html\n",
        "\n",
        "# install HACNet\n",
        "!pip install HACNet\n",
        "\n",
        " # import predict_pkd function from HACNet package\n",
        "from HACNet.functions import predict_pkd\n",
        "\n",
        "# clone GitHub repository containing parameter files for HAC-Net\n",
        "!git clone https://github.com/gregory-kyro/HAC-Net.git\n",
        "\n",
        "# define xml file containing atomic features\n",
        "elements_xml = '/content/HAC-Net/HACNet/element_features.xml'\n",
        "\n",
        "# define 3D-CNN parameter file\n",
        "cnn_params = '/content/HAC-Net/HACNet/parameter_files/CNN_parameters.pt'\n",
        "\n",
        "# define GCN parameter file\n",
        "gcn0_params = '/content/HAC-Net/HACNet/parameter_files/GCN0_parameters.pt'\n",
        "\n",
        "# define other GCN parameter file\n",
        "gcn1_params = '/content/HAC-Net/HACNet/parameter_files/GCN1_parameters.pt'\n",
        "\n",
        "# define MLP parameter file\n",
        "mlp_params = '/content/HAC-Net/HACNet/parameter_files/MLP_parameters.pt'"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#@title Inference Function\n",
        "\n",
        "from biopandas.pdb import PandasPdb\n",
        "from biopandas.mol2 import PandasMol2\n",
        "import numpy as np\n",
        "from pymol import cmd\n",
        "import requests\n",
        "import xml.etree.ElementTree as ET\n",
        "import openbabel.pybel\n",
        "import math\n",
        "import os\n",
        "import torch\n",
        "from torch import Tensor\n",
        "import torch.nn as nn\n",
        "from torch_geometric.nn.conv import MessagePassing, GatedGraphConv\n",
        "from torch_geometric.nn import global_add_pool\n",
        "from torch_geometric.utils import add_self_loops, dense_to_sparse\n",
        "from torch_geometric.nn.aggr import AttentionalAggregation\n",
        "from torch._C import NoneType\n",
        "from torch.optim import Adam\n",
        "from torch_geometric.data import Data, DataListLoader\n",
        "from torch_geometric.nn import DataParallel as GeometricDataParallel\n",
        "from torch.utils.data import Dataset\n",
        "import numpy as np\n",
        "import h5py\n",
        "from sklearn.metrics import r2_score, mean_absolute_error, mean_squared_error, pairwise_distances\n",
        "import os\n",
        "import random\n",
        "from scipy import stats\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "from IPython.display import Image, display\n",
        "\n",
        "os.chdir('/content/HAC-Net/HACNet')\n",
        "\n",
        "from CNN import *\n",
        "from GCN import *\n",
        "\n",
        "\"\"\" define a function to predict pkd for a single protein-ligand complex \"\"\"\n",
        "def predict_pkd(protein_pdb, ligand_mol2, elements_xml, cnn_params, gcn0_params, gcn1_params, mlp_params, verbose=True):\n",
        "\n",
        "  \"\"\"\n",
        "  Inputs:\n",
        "  1) protein_pdb: path to protein pdb file\n",
        "  2) ligand_mol2: path to ligand mol2 file\n",
        "  3) verbose: if True, will return pkd and image of protein-ligand complex.\n",
        "      If False, will return float corresponding to predicted pkd. Default is True\n",
        "\n",
        "  Output:\n",
        "  1) Prediction of pkd\n",
        "  2) visual of protein-ligand complex (if verbose == True)\n",
        "  \"\"\"\n",
        "\n",
        "  \"\"\" define function to extract pocket from protein pdb \"\"\"\n",
        "  def extract_pocket(protein_pdb, ligand_mol2):\n",
        "\n",
        "    # read in protein pdb file\n",
        "    protein =  PandasPdb().read_pdb(protein_pdb)\n",
        "\n",
        "    # define protein atoms dataframe\n",
        "    protein_atom = protein.df['ATOM'].reset_index(drop=True)\n",
        "\n",
        "    # create protein atom dictionary\n",
        "    protein_atom_dict = protein_atom.to_dict('index')\n",
        "\n",
        "    # define protein heteroatoms dataframe\n",
        "    protein_hetatm =  protein.df['HETATM'].reset_index(drop=True)\n",
        "\n",
        "    # create protein heteroatom dictionary\n",
        "    protein_hetatm_dict = protein_hetatm.to_dict('index')\n",
        "\n",
        "    # read in ligand mol2 file\n",
        "    ligand = PandasMol2().read_mol2(ligand_mol2).df\n",
        "\n",
        "    # define ligand non-H atoms dataframe\n",
        "    ligand_nonh = ligand[ligand['atom_type'] != 'H'].reset_index(drop=True)\n",
        "\n",
        "    # create ligand non-H atom dictionary\n",
        "    ligand_nonh_dict = ligand_nonh.to_dict('index')\n",
        "\n",
        "    # initialize lists to save IDs for residues and heteroatoms to keep in pocket file\n",
        "    pocket_residues = []\n",
        "    pocket_heteroatoms = []\n",
        "\n",
        "    # for each protein atom:\n",
        "    for j in range(len(protein_atom_dict)):\n",
        "\n",
        "      # if residue number is not already saved in list\n",
        "      if protein_atom_dict[j]['residue_number'] not in pocket_residues:\n",
        "\n",
        "        # for each ligand non-H atom:\n",
        "        for i in range(len(ligand_nonh_dict)):\n",
        "\n",
        "          # if Euclidean distance is within 8 Angstroms:\n",
        "          if np.sqrt((ligand_nonh_dict[i]['x'] - protein_atom_dict[j]['x_coord'])**2 + (ligand_nonh_dict[i]['y'] - protein_atom_dict[j]['y_coord'])**2 + (ligand_nonh_dict[i]['z'] - protein_atom_dict[j]['z_coord'])**2) <= 8:\n",
        "\n",
        "            # save chain ID, residue number and insertion to list\n",
        "            pocket_residues.append(str(protein_atom_dict[j]['chain_id'])+'_'+str(protein_atom_dict[j]['residue_number'])+'_'+str(protein_atom_dict[j]['insertion']))\n",
        "\n",
        "            break\n",
        "\n",
        "    # for each protein heteroatom:\n",
        "    for k in range(len(protein_hetatm_dict)):\n",
        "\n",
        "      # if residue number is not already saved in list\n",
        "      if protein_hetatm_dict[k]['residue_number'] not in pocket_heteroatoms:\n",
        "\n",
        "        # for each ligand non-H atom:\n",
        "        for i in range(len(ligand_nonh_dict)):\n",
        "\n",
        "          # if Euclidean distance is within 8 Angstroms:\n",
        "          if np.sqrt((ligand_nonh_dict[i]['x'] - protein_hetatm_dict[k]['x_coord'])**2 + (ligand_nonh_dict[i]['y'] - protein_hetatm_dict[k]['y_coord'])**2 + (ligand_nonh_dict[i]['z'] - protein_hetatm_dict[k]['z_coord'])**2) <= 8:\n",
        "\n",
        "            # save heteroatom ID to list\n",
        "            pocket_heteroatoms.append(protein_hetatm_dict[k]['residue_number'])\n",
        "\n",
        "            break\n",
        "\n",
        "    # initialize list to store atoms that are included in saved residues\n",
        "    atoms_to_keep = []\n",
        "\n",
        "    # loop through atom dictionary\n",
        "    for k,v in protein_atom_dict.items():\n",
        "\n",
        "      # if atom is in saved list\n",
        "      if str(v['chain_id'])+'_'+str(v['residue_number'])+'_'+str(v['insertion']) in pocket_residues:\n",
        "\n",
        "        # append the atom number to new list\n",
        "        atoms_to_keep.append(v['atom_number'])\n",
        "\n",
        "    # define the atoms to include in pocket\n",
        "    residues = protein_atom[(protein_atom['atom_number'].isin(atoms_to_keep))]\n",
        "\n",
        "    # reset atom number ordering\n",
        "    residues = residues.reset_index(drop=1)\n",
        "\n",
        "    # initialize list to store heteroatoms that are included in saved heteroatom IDs\n",
        "    hetatms_to_keep = []\n",
        "\n",
        "    # loop through heteroatom dictionary\n",
        "    for k,v in protein_hetatm_dict.items():\n",
        "\n",
        "      # if heteroatom is in saved list\n",
        "      if v['residue_number'] in pocket_heteroatoms and v['residue_name']=='HOH':\n",
        "\n",
        "        # append the heteroatom number to new list\n",
        "        hetatms_to_keep.append(v['atom_number'])\n",
        "\n",
        "    # define the heteroatoms to include in pocket file\n",
        "    heteroatoms = protein_hetatm[(protein_hetatm['atom_number'].isin(hetatms_to_keep))]\n",
        "\n",
        "    # reset heteroatom number ordering\n",
        "    heteroatoms = heteroatoms.reset_index(drop=1)\n",
        "\n",
        "    # initialize biopandas object to write out pocket pdb file\n",
        "    pred_pocket = PandasPdb()\n",
        "\n",
        "    # define the atoms and heteroatoms of the object\n",
        "    pred_pocket.df['ATOM'], pred_pocket.df['HETATM'] = residues, heteroatoms\n",
        "\n",
        "    # save the created object to a pdb file\n",
        "    return pred_pocket.to_pdb('/content/pocket.pdb')\n",
        "\n",
        "  # run function to extract the pocket from protein pdb file\n",
        "  extract_pocket(protein_pdb, ligand_mol2)\n",
        "\n",
        "  # initialize a pymol state by first deleting everything\n",
        "  cmd.delete('all')\n",
        "\n",
        "  # load in the created pocket pdb file\n",
        "  cmd.load('/content/pocket.pdb')\n",
        "\n",
        "  # add hydrogens to the water molecules\n",
        "  cmd.h_add('sol')\n",
        "\n",
        "  # save the state as a mol2 file\n",
        "  cmd.save('/content/pocket.mol2')\n",
        "\n",
        "  \"\"\" define function to calculate charges for pocket mol2 file \"\"\"\n",
        "  def add_mol2_charges(pocket_mol2):\n",
        "\n",
        "    # upload the pocket mol2 file to the ACC2 API\n",
        "    r=requests.post('https://acc2-api.biodata.ceitec.cz/send_files', files={'file[]':open(pocket_mol2,'rb')})\n",
        "\n",
        "    # obtain ID number for uploaded file\n",
        "    r_id=list(r.json()['structure_ids'].values())[0]\n",
        "\n",
        "    # calculate charges using eqeq method\n",
        "    r_out=requests.get('https://acc2-api.biodata.ceitec.cz/calculate_charges?structure_id='+r_id+'&method=eqeq&generate_mol2=true')\n",
        "\n",
        "    # save output mol2 file\n",
        "    open('/content/charged_pocket.mol2', 'wb').write(r_out.content)\n",
        "\n",
        "  # run function to calculate and add charges to pocket mol2 file\n",
        "  add_mol2_charges('/content/pocket.mol2')\n",
        "\n",
        "  # define charged pocket mol2 variable\n",
        "  pocket_mol2_charged = '/content/charged_pocket.mol2'\n",
        "\n",
        "  \"\"\" define Featurizer class from tfbio source code \"\"\"\n",
        "  class Featurizer():\n",
        "\n",
        "      def __init__(self, atom_codes=None, atom_labels=None,\n",
        "                  named_properties=None, save_molecule_codes=True,\n",
        "                  custom_properties=None, smarts_properties=None,\n",
        "                  smarts_labels=None):\n",
        "\n",
        "          # initialize list to store names of all features in the correct order\n",
        "          self.FEATURE_NAMES = []\n",
        "\n",
        "          # validate and process atom codes and labels\n",
        "          if atom_codes is not None:\n",
        "              if not isinstance(atom_codes, dict):\n",
        "                  raise TypeError('Atom codes should be dict, got %s instead'\n",
        "                                  % type(atom_codes))\n",
        "\n",
        "              codes = set(atom_codes.values())\n",
        "              for i in range(len(codes)):\n",
        "                  if i not in codes:\n",
        "                      raise ValueError('Incorrect atom code %s' % i)\n",
        "\n",
        "              self.NUM_ATOM_CLASSES = len(codes)\n",
        "              self.ATOM_CODES = atom_codes\n",
        "\n",
        "              if atom_labels is not None:\n",
        "                  if len(atom_labels) != self.NUM_ATOM_CLASSES:\n",
        "                      raise ValueError('Incorrect number of atom labels: '\n",
        "                                      '%s instead of %s'\n",
        "                                      % (len(atom_labels), self.NUM_ATOM_CLASSES))\n",
        "\n",
        "              else:\n",
        "                  atom_labels = ['atom%s' % i for i in range(self.NUM_ATOM_CLASSES)]\n",
        "\n",
        "              self.FEATURE_NAMES += atom_labels\n",
        "\n",
        "          else:\n",
        "              self.ATOM_CODES = {}\n",
        "              metals = ([3, 4, 11, 12, 13] + list(range(19, 32)) + list(range(37, 51)) + list(range(55, 84)) + list(range(87, 104)))\n",
        "\n",
        "              # List of tuples (atomic_num, class_name) with atom types to encode\n",
        "              atom_classes = [(5, 'B'), (6, 'C'), (7, 'N'), (8, 'O'), (15, 'P'),\n",
        "                  (16, 'S'), (34, 'Se'), ([9, 17, 35, 53], 'halogen'), (metals, 'metal')]\n",
        "\n",
        "              for code, (atom, name) in enumerate(atom_classes):\n",
        "                  if type(atom) is list:\n",
        "                      for a in atom:\n",
        "                          self.ATOM_CODES[a] = code\n",
        "\n",
        "                  else:\n",
        "                      self.ATOM_CODES[atom] = code\n",
        "\n",
        "                  self.FEATURE_NAMES.append(name)\n",
        "\n",
        "              self.NUM_ATOM_CLASSES = len(atom_classes)\n",
        "\n",
        "          # validate and process named properties\n",
        "          if named_properties is not None:\n",
        "              if not isinstance(named_properties, (list, tuple, np.ndarray)):\n",
        "                  raise TypeError('named_properties must be a list')\n",
        "\n",
        "              allowed_props = [prop for prop in dir(openbabel.pybel.Atom)\n",
        "                              if not prop.startswith('__')]\n",
        "\n",
        "              for prop_id, prop in enumerate(named_properties):\n",
        "                  if prop not in allowed_props:\n",
        "                      raise ValueError(\n",
        "                          'named_properties must be in pybel.Atom attributes,'\n",
        "                          ' %s was given at position %s' % (prop_id, prop))\n",
        "\n",
        "              self.NAMED_PROPS = named_properties\n",
        "\n",
        "          else:\n",
        "              # pybel.Atom properties to save\n",
        "              self.NAMED_PROPS = ['hyb', 'heavydegree', 'heterodegree',\n",
        "                                  'partialcharge']\n",
        "\n",
        "          self.FEATURE_NAMES += self.NAMED_PROPS\n",
        "\n",
        "          if not isinstance(save_molecule_codes, bool):\n",
        "              raise TypeError('save_molecule_codes should be bool, got %s '\n",
        "                              'instead' % type(save_molecule_codes))\n",
        "\n",
        "          self.save_molecule_codes = save_molecule_codes\n",
        "\n",
        "          if save_molecule_codes:\n",
        "              # Remember if an atom belongs to the ligand or to the protein\n",
        "              self.FEATURE_NAMES.append('molcode')\n",
        "\n",
        "          # process custom callable properties\n",
        "          self.CALLABLES = []\n",
        "\n",
        "          if custom_properties is not None:\n",
        "              for i, func in enumerate(custom_properties):\n",
        "                  if not callable(func):\n",
        "                      raise TypeError('custom_properties should be list of'\n",
        "                                      ' callables, got %s instead' % type(func))\n",
        "\n",
        "                  name = getattr(func, '__name__', '')\n",
        "\n",
        "                  if name == '':\n",
        "                      name = 'func%s' % i\n",
        "\n",
        "                  self.CALLABLES.append(func)\n",
        "\n",
        "                  self.FEATURE_NAMES.append(name)\n",
        "\n",
        "          # process SMARTS properties and labels\n",
        "          if smarts_properties is None:\n",
        "              # SMARTS definition for other properties\n",
        "              self.SMARTS = [\n",
        "                  '[#6+0!$(*~[#7,#8,F]),SH0+0v2,s+0,S^3,Cl+0,Br+0,I+0]',\n",
        "                  '[a]',\n",
        "                  '[!$([#1,#6,F,Cl,Br,I,o,s,nX3,#7v5,#15v5,#16v4,#16v6,*+1,*+2,*+3])]',\n",
        "                  '[!$([#6,H0,-,-2,-3]),$([!H0;#7,#8,#9])]',\n",
        "                  '[r]']\n",
        "\n",
        "              smarts_labels = ['hydrophobic', 'aromatic', 'acceptor', 'donor',\n",
        "                              'ring']\n",
        "\n",
        "          elif not isinstance(smarts_properties, (list, tuple, np.ndarray)):\n",
        "              raise TypeError('smarts_properties must be a list')\n",
        "\n",
        "          else:\n",
        "              self.SMARTS = smarts_properties\n",
        "\n",
        "          if smarts_labels is not None:\n",
        "              if len(smarts_labels) != len(self.SMARTS):\n",
        "                  raise ValueError('Incorrect number of SMARTS labels: %s'\n",
        "                                  ' instead of %s'\n",
        "                                  % (len(smarts_labels), len(self.SMARTS)))\n",
        "\n",
        "          else:\n",
        "              smarts_labels = ['smarts%s' % i for i in range(len(self.SMARTS))]\n",
        "\n",
        "          # Compile SMARTS patterns for matching\n",
        "          self.compile_smarts()\n",
        "\n",
        "          self.FEATURE_NAMES += smarts_labels\n",
        "\n",
        "      # define function to compile SMARTS patterns for efficient matching\n",
        "      def compile_smarts(self):\n",
        "          self.__PATTERNS = []\n",
        "\n",
        "          for smarts in self.SMARTS:\n",
        "              self.__PATTERNS.append(openbabel.pybel.Smarts(smarts))\n",
        "\n",
        "      # define function to encode the atomic number using one-hot encoding\n",
        "      def encode_num(self, atomic_num):\n",
        "\n",
        "          if not isinstance(atomic_num, int):\n",
        "              raise TypeError('Atomic number must be int, %s was given'\n",
        "                              % type(atomic_num))\n",
        "\n",
        "          encoding = np.zeros(self.NUM_ATOM_CLASSES)\n",
        "\n",
        "          try:\n",
        "              encoding[self.ATOM_CODES[atomic_num]] = 1.0\n",
        "\n",
        "          except:\n",
        "              pass\n",
        "\n",
        "          return encoding\n",
        "\n",
        "      # define function to find substructures in the molecule that match the SMARTS patterns\n",
        "      def find_smarts(self, molecule):\n",
        "\n",
        "          if not isinstance(molecule, openbabel.pybel.Molecule):\n",
        "              raise TypeError('molecule must be pybel.Molecule object, %s was given'\n",
        "                              % type(molecule))\n",
        "\n",
        "          features = np.zeros((len(molecule.atoms), len(self.__PATTERNS)))\n",
        "\n",
        "          for (pattern_id, pattern) in enumerate(self.__PATTERNS):\n",
        "              atoms_with_prop = np.array(list(*zip(*pattern.findall(molecule))),\n",
        "                                          dtype=int) - 1\n",
        "\n",
        "              features[atoms_with_prop, pattern_id] = 1.0\n",
        "\n",
        "          return features\n",
        "\n",
        "      # define function to extract the features from the molecule\n",
        "      def get_features(self, molecule, molcode=None):\n",
        "\n",
        "          if not isinstance(molecule, openbabel.pybel.Molecule):\n",
        "              raise TypeError('molecule must be pybel.Molecule object,'\n",
        "                              ' %s was given' % type(molecule))\n",
        "\n",
        "          if molcode is None:\n",
        "              if self.save_molecule_codes is True:\n",
        "                  raise ValueError('save_molecule_codes is set to True,'\n",
        "                                  ' you must specify code for the molecule')\n",
        "\n",
        "          elif not isinstance(molcode, (float, int)):\n",
        "              raise TypeError('motlype must be float, %s was given'\n",
        "                              % type(molcode))\n",
        "\n",
        "          coords = []\n",
        "          features = []\n",
        "          heavy_atoms = []\n",
        "\n",
        "          for i, atom in enumerate(molecule):\n",
        "\n",
        "              # ignore hydrogens and dummy atoms (they have atomicnum set to 0)\n",
        "              if atom.atomicnum > 1:\n",
        "                  heavy_atoms.append(i)\n",
        "\n",
        "                  coords.append(atom.coords)\n",
        "\n",
        "                  features.append(np.concatenate((\n",
        "                      self.encode_num(atom.atomicnum),\n",
        "                      [atom.__getattribute__(prop) for prop in self.NAMED_PROPS],\n",
        "                      [func(atom) for func in self.CALLABLES])))\n",
        "\n",
        "          coords = np.array(coords, dtype=np.float32)\n",
        "\n",
        "          features = np.array(features, dtype=np.float32)\n",
        "\n",
        "          if self.save_molecule_codes:\n",
        "              features = np.hstack((features,\n",
        "                                      molcode * np.ones((len(features), 1))))\n",
        "\n",
        "          features = np.hstack([features,\n",
        "                                  self.find_smarts(molecule)[heavy_atoms]])\n",
        "\n",
        "          if np.isnan(features).any():\n",
        "              raise RuntimeError('Got NaN when calculating features')\n",
        "\n",
        "          return coords, features\n",
        "\n",
        "      # define function to save the Featurizer to a pickle file\n",
        "      def to_pickle(self, fname='featurizer.pkl'):\n",
        "\n",
        "          # patterns can't be pickled, we need to temporarily remove them\n",
        "          patterns = self.__PATTERNS[:]\n",
        "\n",
        "          del self.__PATTERNS\n",
        "\n",
        "          try:\n",
        "              with open(fname, 'wb') as f:\n",
        "                  pickle.dump(self, f)\n",
        "\n",
        "          finally:\n",
        "              self.__PATTERNS = patterns[:]\n",
        "\n",
        "      @staticmethod\n",
        "      def from_pickle(fname):\n",
        "          with open(fname, 'rb') as f:\n",
        "              featurizer = pickle.load(f)\n",
        "\n",
        "          featurizer.compile_smarts()\n",
        "\n",
        "          return featurizer\n",
        "\n",
        "  \"\"\" define function to featurize mol2 files for use in HAC-Net \"\"\"\n",
        "  def prepare_data(pocket_mol2_charged, ligand_mol2, elements_xml):\n",
        "\n",
        "      # define function to extract features from the binding pocket mol2 file\n",
        "      def __get_pocket():\n",
        "            # define pocket from input pocket_mol2 file\n",
        "            pocket = next(openbabel.pybel.readfile('mol2', pocket_mol2_charged))\n",
        "\n",
        "            # obtain pocket coordinates and features\n",
        "            pocket_coords, pocket_features = featurizer.get_features(pocket, molcode=-1)\n",
        "\n",
        "            # obtain Van der Waals radii of pocket atoms\n",
        "            pocket_vdw = parse_mol_vdw(mol=pocket, element_dict=element_dict)\n",
        "\n",
        "            yield (pocket_coords, pocket_features, pocket_vdw)\n",
        "\n",
        "      # define function to extract information from elements_xml file\n",
        "      def parse_element_description(xml_file):\n",
        "          # initialize dictionary to store element chemical information\n",
        "          element_info_dict = {}\n",
        "\n",
        "          # parse and define elements_xml file\n",
        "          element_info_xml = ET.parse(xml_file)\n",
        "\n",
        "          # for each element in file\n",
        "          for element in element_info_xml.iter():\n",
        "\n",
        "              # if 'comment' is keys of element attributes\n",
        "              if \"comment\" in element.attrib.keys():\n",
        "\n",
        "                  #continue\n",
        "                  continue\n",
        "\n",
        "              # if 'comment' not in keys of element attributes\n",
        "              else:\n",
        "                  # save the attribute to the dictionary value to 'number\n",
        "                  element_info_dict[int(element.attrib[\"number\"])] = element.attrib\n",
        "\n",
        "          # return dictionary containing element information\n",
        "          return element_info_dict\n",
        "\n",
        "      # define function to create a list of van der Waals radii for a molecule\n",
        "      def parse_mol_vdw(mol, element_dict):\n",
        "\n",
        "        # initialize list to store Van der Waals radii\n",
        "          vdw_list = []\n",
        "\n",
        "          # for each atom in molecule\n",
        "          for atom in mol.atoms:\n",
        "\n",
        "              # if the atom is not Hydrogen\n",
        "              if int(atom.atomicnum)>=2:\n",
        "\n",
        "                  # append the Van der Waals radius to the list\n",
        "                  vdw_list.append(float(element_dict[atom.atomicnum][\"vdWRadius\"]))\n",
        "\n",
        "         # return the Van der Waals list as a numpy array\n",
        "          return np.asarray(vdw_list)\n",
        "\n",
        "      # read in elements_xml and store important information in dictionary\n",
        "      element_dict = parse_element_description(elements_xml)\n",
        "\n",
        "      # define featurizer object\n",
        "      featurizer = Featurizer()\n",
        "\n",
        "      # define object for getting features of pocket\n",
        "      pocket_generator = __get_pocket()\n",
        "\n",
        "      # read ligand file using pybel\n",
        "      ligand = next(openbabel.pybel.readfile('mol2', ligand_mol2))\n",
        "\n",
        "      # extract coordinates, 19 features, and Van der Waals radii from pocket atoms\n",
        "      pocket_coords, pocket_features, pocket_vdw = next(pocket_generator)\n",
        "\n",
        "      # extract coordinates, and 19 features from ligand atoms\n",
        "      ligand_coords, ligand_features = featurizer.get_features(ligand, molcode=1)\n",
        "\n",
        "      # extract Van der Waals radii from ligand atoms\n",
        "      ligand_vdw = parse_mol_vdw(mol=ligand, element_dict=element_dict)\n",
        "\n",
        "      # define centroid to be the center of the ligand\n",
        "      centroid = ligand_coords.mean(axis=0)\n",
        "\n",
        "      # normalize ligand coordinates with respect to centroid\n",
        "      ligand_coords -= centroid\n",
        "\n",
        "      #normalize pocket coordinates with respect to centroid\n",
        "      pocket_coords -= centroid\n",
        "\n",
        "      # assemble the features into one large numpy array where rows are heavy atoms, and columns are coordinates and features\n",
        "      data = np.concatenate((np.concatenate((ligand_coords, pocket_coords)),\n",
        "          np.concatenate((ligand_features, pocket_features))), axis=1)\n",
        "\n",
        "      # concatenate van der Waals radii into one numpy array\n",
        "      vdw_radii = np.concatenate((ligand_vdw, pocket_vdw))\n",
        "\n",
        "      # return properly formatted coordinates, features, and Van der Waals radii\n",
        "      return data, vdw_radii\n",
        "\n",
        "  # prepare data from input files\n",
        "  prep_data, prep_vdw = prepare_data(pocket_mol2_charged, ligand_mol2, elements_xml)\n",
        "\n",
        "  # define function to voxelize input data for use in 3D-CNN component\n",
        "  def voxelize_one_vox_per_atom(xyz_array, feat, vol_dim):\n",
        "\n",
        "    # initialize volume\n",
        "    vol_data = np.zeros((vol_dim[0], vol_dim[1], vol_dim[2], vol_dim[3]), dtype=np.float32)\n",
        "\n",
        "    # get coordinates of center of x axis\n",
        "    xmid = (min(xyz_array[:, 0]) + max(xyz_array[:, 0])) / 2\n",
        "\n",
        "    # get coordinates of center of y axis\n",
        "    ymid = (min(xyz_array[:, 1]) + max(xyz_array[:, 1])) / 2\n",
        "\n",
        "    # get coordinates of center of z axis\n",
        "    zmid = (min(xyz_array[:, 2]) + max(xyz_array[:, 2])) / 2\n",
        "\n",
        "    # define minimum x coordinate\n",
        "    xmin = xmid - (48 / 2)\n",
        "\n",
        "    # define minimum y coordinate\n",
        "    ymin = ymid - (48 / 2)\n",
        "\n",
        "    # define minimum z coordinate\n",
        "    zmin = zmid - (48 / 2)\n",
        "\n",
        "    # define maximum x coordinate\n",
        "    xmax = xmid + (48 / 2)\n",
        "\n",
        "    # define maximum y coordinate\n",
        "    ymax = ymid + (48 / 2)\n",
        "\n",
        "    # define maximum z coordinate\n",
        "    zmax = zmid + (48 / 2)\n",
        "\n",
        "    # assign each atom to the voxel that contains its center\n",
        "    for ind in range(xyz_array.shape[0]):\n",
        "\n",
        "        # define x coordinate\n",
        "        x = xyz_array[ind, 0]\n",
        "\n",
        "        # define y coordinate\n",
        "        y = xyz_array[ind, 1]\n",
        "\n",
        "        # define z coordinate\n",
        "        z = xyz_array[ind, 2]\n",
        "\n",
        "        # if atom is not within the voxel grid\n",
        "        if x < xmin or x >= xmax or y < ymin or y >= ymax or z < zmin or z >= zmax:\n",
        "\n",
        "            # continue\n",
        "            continue\n",
        "\n",
        "        # determine x-axis index of voxel to contain atom\n",
        "        cx = math.floor((x - xmin) / (xmax - xmin) * (vol_dim[1]))\n",
        "\n",
        "        # determine y-axis index of voxel to contain atom\n",
        "        cy = math.floor((y - ymin) / (ymax - ymin) * (vol_dim[2]))\n",
        "\n",
        "        # determine z-axis index of voxel to contain atom\n",
        "        cz = math.floor((z - zmin) / (zmax - zmin) * (vol_dim[3]))\n",
        "\n",
        "        # add to each voxel the features of the assigned atom\n",
        "        vol_data[:, cx, cy, cz] += feat[ind, :]\n",
        "\n",
        "    # return voxelized data\n",
        "    return vol_data\n",
        "\n",
        "  # prepare voxelized data\n",
        "  prep_data_vox = voxelize_one_vox_per_atom(prep_data[:, 0:3], prep_data[:, 3:], [prep_data.shape[1]-3, 48, 48, 48])\n",
        "\n",
        "  # add an additional axis and convert to a tensor\n",
        "  prep_data_vox = torch.tensor(prep_data_vox[np.newaxis,...])\n",
        "\n",
        "\n",
        "  \"\"\" define a function to extract flattened features from trained 3D-CNN \"\"\"\n",
        "  def extract_features(checkpoint_path):\n",
        "\n",
        "      # set CUDA for PyTorch\n",
        "      use_cuda = torch.cuda.is_available()\n",
        "      cuda_count = torch.cuda.device_count()\n",
        "\n",
        "      # if using GPU\n",
        "      if use_cuda:\n",
        "          device_name = \"cuda:0\"\n",
        "          device = torch.device(device_name)\n",
        "          torch.cuda.set_device(int(device_name.split(':')[1]))\n",
        "\n",
        "      # if not using GPU\n",
        "      else:\n",
        "          # use CPU\n",
        "          device = torch.device(\"cpu\")\n",
        "\n",
        "      # define model\n",
        "      model = CNN(use_cuda=use_cuda)\n",
        "      model.to(device)\n",
        "\n",
        "      # load checkpoint file\n",
        "      checkpoint = torch.load(checkpoint_path, map_location=device)\n",
        "\n",
        "      # prepare model state dict from CNN parameter file\n",
        "      model_state_dict = checkpoint.pop(\"model_state_dict\")\n",
        "\n",
        "      # load state dict\n",
        "      model.load_state_dict(model_state_dict, strict=False)\n",
        "\n",
        "      # put model in evaluation mode\n",
        "      model.eval()\n",
        "\n",
        "      with torch.no_grad():\n",
        "\n",
        "          # define flattened features\n",
        "          x_batch_cpu = prep_data_vox\n",
        "          x_batch = x_batch_cpu.to(device)\n",
        "          _, flat_feat_batch = model(x_batch)\n",
        "          flatfeat = flat_feat_batch.cpu().data.numpy()\n",
        "\n",
        "      # return flattened features\n",
        "      return flatfeat\n",
        "\n",
        "  # define extracted features for training MLP\n",
        "  flat_feat = extract_features(checkpoint_path = cnn_params)\n",
        "\n",
        "  \"\"\" define function to perform forward pass and return prediction \"\"\"\n",
        "  def test_HACNet(cnn_params, gcn0_params, gcn1_params):\n",
        "\n",
        "      # set CUDA for PyTorch\n",
        "      use_cuda = torch.cuda.is_available()\n",
        "      cuda_count = torch.cuda.device_count()\n",
        "      device_name = \"cuda:0\"\n",
        "\n",
        "      # if using GPU\n",
        "      if use_cuda:\n",
        "          device = torch.device(device_name)\n",
        "          torch.cuda.set_device(int(device_name.split(':')[1]))\n",
        "\n",
        "      # if not using GPU\n",
        "      else:\n",
        "\n",
        "          # use CPU\n",
        "          device = torch.device('cpu')\n",
        "\n",
        "      # define 3D-CNN (MLP) model\n",
        "      cnn_model = Model_Linear(use_cuda=use_cuda)\n",
        "      cnn_model.to(device)\n",
        "\n",
        "      # load checkpoint file\n",
        "      cnn_checkpoint = torch.load(cnn_params, map_location=device)\n",
        "\n",
        "      # load in parameters and fill model state dict\n",
        "      cnn_model_state_dict = cnn_checkpoint.pop(\"model_state_dict\")\n",
        "      cnn_model.load_state_dict(cnn_model_state_dict, strict=False)\n",
        "\n",
        "      # put model in evaluation mode\n",
        "      cnn_model.eval()\n",
        "\n",
        "      with torch.no_grad():\n",
        "\n",
        "          # obtain 3D-CNN (MLP) prediction from flattened feature input\n",
        "          x_batch_cpu = torch.tensor(flat_feat)\n",
        "          x_batch = x_batch_cpu.to(device)\n",
        "          ypred_batch, _ = cnn_model(x_batch)\n",
        "          ypred = ypred_batch.cpu().float().data.numpy()[:,0]\n",
        "          y_pred_cnn = ypred\n",
        "\n",
        "      # define first GCN model (GCN0)\n",
        "      gcn0_model = GeometricDataParallel(GCN(in_channels=20, gather_width=128, prop_iter=4, dist_cutoff=3.5)).float()\n",
        "\n",
        "      # load checkpoint file\n",
        "      gcn0_checkpoint = torch.load(gcn0_params, map_location=device)\n",
        "\n",
        "      # fill model state dict with parameters\n",
        "      gcn0_model_state_dict = gcn0_checkpoint.pop(\"model_state_dict\")\n",
        "      gcn0_model.load_state_dict(gcn0_model_state_dict, strict=False)\n",
        "\n",
        "      # put model in evaluation mode\n",
        "      gcn0_model.eval()\n",
        "\n",
        "      # initialize list to store predictions\n",
        "      y_pred_gcn0 = []\n",
        "\n",
        "      with torch.no_grad():\n",
        "\n",
        "        # obtain predictions from input data\n",
        "        data_list = []\n",
        "        vdw_radii = prep_vdw.reshape(-1, 1)\n",
        "        node_feats = np.concatenate([vdw_radii,prep_data[:, 3:22]], axis=1)\n",
        "        coords=prep_data[:,0:3]\n",
        "        dists=pairwise_distances(coords, metric='euclidean')\n",
        "        edge_index, edge_attr = dense_to_sparse(torch.from_numpy(dists).float())\n",
        "        x = torch.from_numpy(node_feats).float()\n",
        "        data = Data(x=x, edge_index=edge_index, edge_attr=edge_attr.view(-1, 1))\n",
        "        data_list.append(data)\n",
        "        batch_data = [x for x in data_list if x is not None]\n",
        "        y_ = gcn0_model(batch_data).cpu().data.numpy()\n",
        "\n",
        "      # define predictions of first GCN\n",
        "      y_pred_gcn0 = np.concatenate(y_).reshape(-1, 1).squeeze(1)\n",
        "\n",
        "      # define second GCN model (GCN1)\n",
        "      gcn1_model = GeometricDataParallel(GCN(in_channels=20, gather_width=128, prop_iter=4, dist_cutoff=3.5)).float()\n",
        "\n",
        "      # load checkpoint file\n",
        "      gcn1_checkpoint = torch.load(gcn1_params, map_location=device)\n",
        "\n",
        "      # fill model state dict with parameters\n",
        "      gcn1_model_state_dict = gcn1_checkpoint.pop(\"model_state_dict\")\n",
        "      gcn1_model.load_state_dict(gcn1_model_state_dict, strict=False)\n",
        "\n",
        "      # put second GCN in evaluation mode\n",
        "      gcn1_model.eval()\n",
        "\n",
        "      # initialize list to store predictions\n",
        "      y_pred_gcn1 = []\n",
        "\n",
        "      with torch.no_grad():\n",
        "\n",
        "        # obtain predictions from input data\n",
        "        data_list = []\n",
        "        vdw_radii = prep_vdw.reshape(-1, 1)\n",
        "        node_feats = np.concatenate([vdw_radii,prep_data[:, 3:22]], axis=1)\n",
        "        coords=prep_data[:,0:3]\n",
        "        dists=pairwise_distances(coords, metric='euclidean')\n",
        "        edge_index, edge_attr = dense_to_sparse(torch.from_numpy(dists).float())\n",
        "        x = torch.from_numpy(node_feats).float()\n",
        "        data = Data(x=x, edge_index=edge_index, edge_attr=edge_attr.view(-1, 1))\n",
        "        data_list.append(data)\n",
        "        batch_data = [x for x in data_list if x is not None]\n",
        "        y_ = gcn1_model(batch_data).cpu().data.numpy()\n",
        "\n",
        "      # define predictions of second GCN\n",
        "      y_pred_gcn1 = np.concatenate(y_).reshape(-1, 1).squeeze(1)\n",
        "\n",
        "      # in the very rare case that the 3D-CNN provides an unreasonable prediction (~ 1 in 1000 PDBbind complexes):\n",
        "      if abs(y_pred_cnn) >= 50:\n",
        "\n",
        "        # return the average prediction of the two GCN models\n",
        "        y_pred = y_pred_gcn0/2 + y_pred_gcn1/2\n",
        "\n",
        "      # if 3D-CNN provides a reasonable prediction (over 99.9% of time)\n",
        "      else:\n",
        "\n",
        "        # return the prediction of binding affinity from complete model\n",
        "        y_pred = y_pred_cnn/3 + y_pred_gcn0/3 + y_pred_gcn1/3\n",
        "\n",
        "      # return prediction\n",
        "      return y_pred[0]\n",
        "\n",
        "  # define predicted binding affinity (pKd)\n",
        "  pkd = test_HACNet(cnn_params = mlp_params,\n",
        "            gcn0_params = gcn0_params,\n",
        "            gcn1_params = gcn1_params)\n",
        "\n",
        "  # remove intermediate files\n",
        "  os.remove('/content/pocket.pdb')\n",
        "  os.remove('/content/pocket.mol2')\n",
        "  os.remove('/content/charged_pocket.mol2')\n",
        "\n",
        "  # if verbose is False\n",
        "  if verbose ==False:\n",
        "\n",
        "    # return only pKd\n",
        "    return round(pkd, 3)\n",
        "\n",
        "  # if verbose is True\n",
        "  if verbose == True:\n",
        "\n",
        "    # return pkd and visual of protein-ligand complex\n",
        "    print('Predicted binding affinity:', str(round(pkd, 3))+' pKd')\n",
        "\n",
        "    # initialize PyMOL session by deleting everything\n",
        "    cmd.delete('all')\n",
        "\n",
        "    # load in ligand twice for representing it visually two different ways\n",
        "    cmd.load(ligand_mol2, 'ligand')\n",
        "    cmd.load(ligand_mol2, 'ligand2')\n",
        "\n",
        "    # load in initial protein file\n",
        "    cmd.load(protein_pdb, 'pocket')\n",
        "\n",
        "    # remove waters\n",
        "    cmd.remove('sol')\n",
        "\n",
        "    # color the ligand and protein\n",
        "    cmd.color('lime', 'ligand2')\n",
        "    cmd.color('violetpurple', 'pocket')\n",
        "    cmd.color('limon', 'ligand')\n",
        "\n",
        "    # show protein as cartoon\n",
        "    cmd.show_as('cartoon', 'pocket')\n",
        "\n",
        "    # set size of spheres to represent ligand atoms\n",
        "    cmd.set('sphere_scale', 0.3)\n",
        "\n",
        "    # show ligand as surface\n",
        "    cmd.show_as('surface', 'ligand2')\n",
        "\n",
        "    # set transparencies\n",
        "    cmd.set('transparency', 0.55, 'ligand2')\n",
        "    cmd.set('transparency', 0.2, 'pocket')\n",
        "\n",
        "    # zoom in on the ligand\n",
        "    cmd.zoom('ligand', 9)\n",
        "\n",
        "    # save session as image\n",
        "    cmd.png('image.png')\n",
        "\n",
        "    # display the image\n",
        "    display(Image('image.png', unconfined=True))\n",
        "\n",
        "\n",
        "\"\"\" Define fully-connected network class \"\"\"\n",
        "class Model_Linear(nn.Module):\n",
        "\n",
        "\tdef __init__(self, use_cuda=True):\n",
        "\t\tsuper(Model_Linear, self).__init__()\n",
        "\t\tself.use_cuda = use_cuda\n",
        "\t\tself.fc1 = nn.Linear(2048, 100)\n",
        "\t\ttorch.nn.init.normal_(self.fc1.weight, 0, 1)\n",
        "\t\tself.fc1_bn = nn.BatchNorm1d(num_features=100, affine=True, momentum=0.3).train()\n",
        "\t\tself.fc2 = nn.Linear(100, 1)\n",
        "\t\ttorch.nn.init.normal_(self.fc2.weight, 0, 1)\n",
        "\t\tself.relu = nn.ReLU()\n",
        "\n",
        "\tdef forward(self, x):\n",
        "\t\tfc1_z = self.fc1(x)\n",
        "\t\tfc1_y = self.relu(fc1_z)\n",
        "\t\tfc1 = self.fc1_bn(fc1_y) if fc1_y.shape[0]>1 else fc1_y\n",
        "\t\tfc2_z = self.fc2(fc1)\n",
        "\t\treturn fc2_z, fc1_z\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "class GCN(torch.nn.Module):\n",
        "\n",
        "    def __init__(self, in_channels, gather_width=128, prop_iter=4, dist_cutoff=3.5):\n",
        "        super(GCN, self).__init__()\n",
        "\n",
        "        # define distance cutoff\n",
        "        self.dist_cutoff=torch.Tensor([dist_cutoff])\n",
        "        if torch.cuda.is_available():\n",
        "            self.dist_cutoff = self.dist_cutoff.cuda()\n",
        "\n",
        "        # attentional aggregation\n",
        "        self.gate_net = nn.Sequential(nn.Linear(in_channels, int(in_channels/2)), nn.Softsign(), nn.Linear(int(in_channels/2), int(in_channels/4)), nn.Softsign(), nn.Linear(int(in_channels/4),1))\n",
        "        self.attn_aggr = AttentionalAggregation(self.gate_net)\n",
        "\n",
        "        # Gated Graph Neural Network\n",
        "        self.gate = GatedGraphConv(in_channels, prop_iter, aggr = self.attn_aggr)\n",
        "\n",
        "        # simple neural networks for use in asymmetric attentional aggregation\n",
        "        self.attn_net_i=nn.Sequential(nn.Linear(in_channels * 2, in_channels), nn.Softsign(),nn.Linear(in_channels, gather_width), nn.Softsign())\n",
        "        self.attn_net_j=nn.Sequential(nn.Linear(in_channels, gather_width), nn.Softsign())\n",
        "\n",
        "        # final set of linear layers for making affinity prediction\n",
        "        self.output = nn.Sequential(nn.Linear(gather_width, int(gather_width / 1.5)), nn.ReLU(), nn.Linear(int(gather_width / 1.5), int(gather_width / 2)), nn.ReLU(), nn.Linear(int(gather_width / 2), 1))\n",
        "\n",
        "    def forward(self, data):\n",
        "\n",
        "        # move data to GPU\n",
        "        if torch.cuda.is_available():\n",
        "            data.x = data.x.cuda()\n",
        "            data.edge_attr = data.edge_attr.cuda()\n",
        "            data.edge_index = data.edge_index.cuda()\n",
        "            data.batch = data.batch.cuda()\n",
        "\n",
        "        # allow nodes to propagate messages to themselves\n",
        "        data.edge_index, data.edge_attr = add_self_loops(data.edge_index, data.edge_attr.view(-1))\n",
        "\n",
        "        # restrict edges to the distance cutoff\n",
        "        row, col = data.edge_index\n",
        "        mask = data.edge_attr <= self.dist_cutoff\n",
        "        mask = mask.squeeze()\n",
        "        row, col, edge_feat = row[mask], col[mask], data.edge_attr[mask]\n",
        "        edge_index=torch.stack([row,col],dim=0)\n",
        "\n",
        "        # propagation\n",
        "        node_feat_0 = data.x\n",
        "        node_feat_1 = self.gate(node_feat_0, edge_index, edge_feat)\n",
        "        node_feat_attn = torch.nn.Softmax(dim=1)(self.attn_net_i(torch.cat([node_feat_1, node_feat_0], dim=1))) * self.attn_net_j(node_feat_0)\n",
        "\n",
        "        # globally sum features and apply linear layers\n",
        "        pool_x = global_add_pool(node_feat_attn, data.batch)\n",
        "        prediction = self.output(pool_x)\n",
        "\n",
        "        return prediction"
      ],
      "metadata": {
        "cellView": "form",
        "id": "m5LV9-0pv39G"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#@title Predict Binding Affinity (pKd)\n",
        "#@markdown **Enter the paths to the protein and ligand structural files, select\n",
        "#@markdown the type of output you wish to receive, and then press the *Play*\n",
        "#@markdown button on the left.**\n",
        "\n",
        "import warnings\n",
        "warnings.filterwarnings(\"ignore\")\n",
        "\n",
        "protein = ''  #@param {type:'string'}\n",
        "ligand = ''  #@param {type:'string'}\n",
        "output = \"only pKd\" #@param [\"pKd and visual\", \"only pKd\"] {allow-input: true}\n",
        "\n",
        "if output == \"pKd and visual\":\n",
        "  verbose = True\n",
        "elif output == 'only pKd':\n",
        "  verbose = False\n",
        "\n",
        "predict_pkd(protein_pdb=protein, ligand_mol2=ligand, elements_xml=elements_xml,\n",
        "            cnn_params=cnn_params, gcn0_params=gcn0_params, gcn1_params=gcn1_params,\n",
        "            mlp_params=mlp_params, verbose=verbose)"
      ],
      "metadata": {
        "id": "3EjmhS9bEE2I",
        "cellView": "form"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}