HinSAGE_test_mol_to_species.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "3ab86b3d",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' \n",
    "import networkx as nx\n",
    "import itertools\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from stellargraph import StellarGraph\n",
    "from tensorflow import keras\n",
    "import tensorflow as tf\n",
    "\n",
    "import stellargraph as sg\n",
    "from stellargraph.data import EdgeSplitter\n",
    "from stellargraph.mapper import HinSAGELinkGenerator\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "from tensorflow.keras import mixed_precision"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "9c8073a0",
   "metadata": {},
   "outputs": [],
   "source": [
    "g = nx.read_graphml(\"./graph/test_graph.gml\")\n",
    "species_features_dummy = pd.read_csv(\"./data/species_features.csv.gz\", index_col=0)\n",
    "molecule_features_dummy = pd.read_csv(\"./data/molecule_features.csv.gz\", index_col=0)\n",
    "df_agg = pd.read_csv(\"./data/lotus_agg_test.csv.gz\", index_col=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "362a0707-3827-4f4f-8eed-018da3795a1d",
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.encoding import binary_encode_df\n",
    "\n",
    "species_features_dummy = binary_encode_df(species_features_dummy)\n",
    "molecule_features_dummy = binary_encode_df(molecule_features_dummy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "d77709dc-7c36-4345-b0ea-f1bf2bac9349",
   "metadata": {},
   "outputs": [],
   "source": [
    "rdkit = pd.read_csv(\"./data/mol_dummy_rdkit.csv.gz\", index_col=0)\n",
    "molecule_features_dummy = molecule_features_dummy.merge(rdkit,\n",
    "                                                        left_index=True,\n",
    "                                                        right_index=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "e8b35e23",
   "metadata": {},
   "outputs": [],
   "source": [
    "species_feat = species_features_dummy[species_features_dummy.index.isin(df_agg.organism_name)]\n",
    "molecule_feat = molecule_features_dummy[molecule_features_dummy.index.isin(df_agg.structure_smiles_2D)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "d2fa5cf0",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_agg_train = pd.read_csv(\"./data/lotus_agg_train.csv.gz\", index_col=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "4972634f",
   "metadata": {},
   "outputs": [],
   "source": [
    "species_unique_to_test_set = df_agg[~df_agg.organism_name.isin(df_agg_train.organism_name)]\n",
    "molecules_unique_to_test_set = df_agg[~df_agg.structure_smiles_2D.isin(df_agg_train.structure_smiles_2D)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "6fefa095",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.2360753087252851"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(molecules_unique_to_test_set)/148190"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "298d505e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "34984"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(molecules_unique_to_test_set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "944a1b71",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3057"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(species_unique_to_test_set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "4f0db4ea",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.08044736842105263"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(species_unique_to_test_set)/38000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "4e68fe2f",
   "metadata": {},
   "outputs": [],
   "source": [
    "g_train = nx.read_graphml(\"./graph/train_graph.gml\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "007a1ea9",
   "metadata": {},
   "outputs": [],
   "source": [
    "g = nx.create_empty_copy(g)\n",
    "g_merged = nx.compose(g_train, g)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "0464b2c6-af7b-43f2-8516-6bfda89746ac",
   "metadata": {},
   "outputs": [],
   "source": [
    "molecule_features_dummy = molecule_features_dummy[~molecule_features_dummy.index.duplicated(keep='first')]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "f424c8fb",
   "metadata": {},
   "outputs": [],
   "source": [
    "#g = nx.create_empty_copy(g)\n",
    "G = StellarGraph.from_networkx(g_merged,\n",
    "                               node_features={'species': species_features_dummy,\n",
    "                                              'molecule':molecule_features_dummy})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "a27b99f5",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "StellarDiGraph: Directed multigraph\n",
      " Nodes: 184194, Edges: 608638\n",
      "\n",
      " Node types:\n",
      "  molecule: [147747]\n",
      "    Features: float32 vector, length 155\n",
      "    Edge types: molecule-present_in->species\n",
      "  species: [36447]\n",
      "    Features: float32 vector, length 69\n",
      "    Edge types: species-has->molecule\n",
      "\n",
      " Edge types:\n",
      "    species-has->molecule: [304319]\n",
      "        Weights: all 1 (default)\n",
      "        Features: none\n",
      "    molecule-present_in->species: [304319]\n",
      "        Weights: all 1 (default)\n",
      "        Features: none\n"
     ]
    }
   ],
   "source": [
    "print(G.info())\n",
    "G.check_graph_for_ml()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "51115395",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = tf.keras.models.load_model(\"./model/gbif_batch_128_layer_1024_m_to_s\", compile=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "6b0ceeb8",
   "metadata": {},
   "outputs": [],
   "source": [
    "species_unique_to_test_set = species_unique_to_test_set[['structure_smiles_2D', 'organism_name']].to_numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "209c7201",
   "metadata": {},
   "outputs": [],
   "source": [
    "def predict(model, flow, iterations=10):\n",
    "    predictions = []\n",
    "    for _ in range(iterations):\n",
    "        predictions.append(model.predict(flow, workers=-1).flatten())\n",
    "\n",
    "    return np.mean(predictions, axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "12d2251f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "24/24 [==============================] - 1s 19ms/step\n",
      "24/24 [==============================] - 0s 16ms/step\n",
      "24/24 [==============================] - 0s 16ms/step\n",
      "24/24 [==============================] - 0s 15ms/step\n",
      "24/24 [==============================] - 0s 17ms/step\n",
      "24/24 [==============================] - 0s 16ms/step\n",
      "24/24 [==============================] - 1s 20ms/step\n",
      "24/24 [==============================] - 0s 18ms/step\n",
      "24/24 [==============================] - 0s 15ms/step\n",
      "24/24 [==============================] - 0s 15ms/step\n"
     ]
    }
   ],
   "source": [
    "result = predict(model, test_flow)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "4224f503",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.741249591102388"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sum(result>0.5)/len(result)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "8b5534f9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'Test on unknown SPECIES but known MOLECULES')"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(result, bins=30)\n",
    "plt.title(\"Test on unknown SPECIES but known MOLECULES\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d4fd1a35-4d2d-4d10-82d8-be1df1ffac7f",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_flow = HinSAGELinkGenerator(G,\n",
    "                                 batch_size=128,\n",
    "                                num_samples=[3,1],\n",
    "                                head_node_types=[\"molecule\", \"species\"]).flow(species_unique_to_test_set,\n",
    "                                                                              np.ones(len(species_unique_to_test_set)).reshape(-1,1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "fd8ab658",
   "metadata": {},
   "outputs": [],
   "source": [
    "test = result[(result>0.9) | (result<0.1)]\n",
    "test = test>0.5\n",
    "test = test.astype('int8')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "aa903743",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "molecules_unique_to_test_set = molecules_unique_to_test_set[['structure_smiles_2D', 'organism_name']].to_numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "83b8cc35",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_flow_mol = HinSAGELinkGenerator(G,\n",
    "                                 batch_size=128,\n",
    "                                num_samples=[3,1],\n",
    "                                head_node_types=[\"molecule\", \"species\"]).flow(molecules_unique_to_test_set,\n",
    "                                                                              np.ones(len(molecules_unique_to_test_set)).reshape(-1,1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "f92f2826",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "274/274 [==============================] - 4s 14ms/step\n",
      "274/274 [==============================] - 4s 15ms/step\n",
      "274/274 [==============================] - 4s 14ms/step\n",
      "274/274 [==============================] - 4s 14ms/step\n",
      "274/274 [==============================] - 4s 15ms/step\n",
      "274/274 [==============================] - 4s 14ms/step\n",
      "274/274 [==============================] - 4s 14ms/step\n",
      "274/274 [==============================] - 4s 14ms/step\n",
      "274/274 [==============================] - 4s 14ms/step\n",
      "274/274 [==============================] - 4s 14ms/step\n"
     ]
    }
   ],
   "source": [
    "out_mol = predict(model, test_flow_mol)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "2921d241",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'Test on unknown MOLECULES but known SPECIES')"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(out_mol, bins=50)\n",
    "plt.title(\"Test on unknown MOLECULES but known SPECIES\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "152c0a20",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9254802195289276"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sum(out_mol>0.5)/len(out_mol)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "cdae65dc",
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.utils import predict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "82b0d764-f6af-4b20-bde8-516ffdf39158",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "molecules_unique_to_test_set = molecules_unique_to_test_set[['structure_smiles_2D', 'organism_name']].copy()\n",
    "molecules_unique_to_test_set.rename(columns={'structure_smiles_2D':'molecule',\n",
    "                                             'organism_name':'species'}, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "7e2cd07a",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_m_to_s = tf.keras.models.load_model(\"./model/gbif_batch_128_layer_1024_m_to_s\", compile=True)\n",
    "model_s_to_m = tf.keras.models.load_model(\"./model/gbif_batch_128_layer_1024_s_to_m\", compile=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "f50a907d-2f70-4f51-a118-e90412da9d84",
   "metadata": {},
   "outputs": [],
   "source": [
    "species_features_dummy = pd.read_csv(\"./data/species_features.csv.gz\", index_col=0)\n",
    "molecule_features_dummy = pd.read_csv(\"./data/molecule_features.csv.gz\", index_col=0)\n",
    "df_agg = pd.read_csv(\"./data/lotus_agg_test.csv.gz\", index_col=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "e1f8b34e-5de3-4ec4-b36a-4a412587622f",
   "metadata": {},
   "outputs": [],
   "source": [
    "rdkit = pd.read_csv(\"./data/mol_dummy_rdkit.csv.gz\", index_col=0).astype('uint8')\n",
    "rdkit.columns = rdkit.columns.astype(str)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "b3cb5453-4b05-4eda-be19-06d222c10739",
   "metadata": {},
   "outputs": [],
   "source": [
    "g_train = nx.read_graphml(\"./graph/train_graph.gml\")\n",
    "g = nx.read_graphml(\"./graph/test_graph.gml\")\n",
    "#g  = nx.create_empty_copy(g)\n",
    "g_merged = nx.compose(g_train, g)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "id": "f2c22fd5-6714-4ef5-b695-de8743b1cf64",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Converting Inchikeys to SMILES...\n",
      "Getting missing features...\n",
      "Converting species taxonomy as numeric...\n",
      "Converting molecules as numeric...\n",
      "Checking which model should be used for each row...\n",
      "Adding missing nodes to the graph...\n",
      "Converting NetworkX to Stellargraph...\n",
      "StellarDiGraph: Directed multigraph\n",
      " Nodes: 184194, Edges: 869484\n",
      "\n",
      " Node types:\n",
      "  molecule: [147747]\n",
      "    Features: float32 vector, length 155\n",
      "    Edge types: molecule-present_in->species\n",
      "  species: [36447]\n",
      "    Features: float32 vector, length 69\n",
      "    Edge types: species-has->molecule\n",
      "\n",
      " Edge types:\n",
      "    species-has->molecule: [434742]\n",
      "        Weights: all 1 (default)\n",
      "        Features: none\n",
      "    molecule-present_in->species: [434742]\n",
      "        Weights: all 1 (default)\n",
      "        Features: none\n",
      "Creating 'forward', 'backward' flow for KNOWN molecule AND species...\n",
      "Predict both : running molecule to species predictions...\n",
      "35/35 [==============================] - 5s 142ms/step\n",
      "35/35 [==============================] - 7s 195ms/step\n",
      "35/35 [==============================] - 5s 140ms/step\n",
      "35/35 [==============================] - 5s 134ms/step\n",
      "35/35 [==============================] - 6s 182ms/step\n",
      "35/35 [==============================] - 8s 220ms/step\n",
      "35/35 [==============================] - 5s 152ms/step\n",
      "Predict both : running species to molecules prediction...\n",
      "35/35 [==============================] - 5s 138ms/step\n",
      "35/35 [==============================] - 5s 146ms/step\n",
      "35/35 [==============================] - 6s 175ms/step\n",
      "35/35 [==============================] - 5s 155ms/step\n",
      "35/35 [==============================] - 5s 138ms/step\n",
      "35/35 [==============================] - 7s 185ms/step\n",
      "35/35 [==============================] - 5s 148ms/step\n"
     ]
    }
   ],
   "source": [
    "out = predict(g_merged,\n",
    "                model_m_to_s,\n",
    "                model_s_to_m,\n",
    "                molecules_unique_to_test_set,\n",
    "                molecule_features_dummy,\n",
    "                rdkit,\n",
    "                species_features_dummy\n",
    "               )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "0e1b54fc-92e2-4327-9d1c-e9aa603e4a54",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([2.000e+00, 0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00,\n",
       "        0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00,\n",
       "        0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00,\n",
       "        0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00, 1.000e+00, 0.000e+00,\n",
       "        1.000e+00, 1.000e+00, 0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00,\n",
       "        0.000e+00, 1.000e+00, 0.000e+00, 1.000e+00, 1.000e+00, 0.000e+00,\n",
       "        1.000e+00, 1.000e+00, 1.000e+00, 0.000e+00, 1.000e+00, 0.000e+00,\n",
       "        1.000e+00, 2.000e+00, 0.000e+00, 0.000e+00, 0.000e+00, 0.000e+00,\n",
       "        2.000e+00, 1.000e+00, 0.000e+00, 0.000e+00, 2.000e+00, 2.000e+00,\n",
       "        1.000e+00, 3.000e+00, 1.000e+00, 0.000e+00, 0.000e+00, 1.000e+00,\n",
       "        3.000e+00, 4.000e+00, 0.000e+00, 2.000e+00, 4.000e+00, 2.000e+00,\n",
       "        3.000e+00, 1.000e+00, 0.000e+00, 4.000e+00, 0.000e+00, 2.000e+00,\n",
       "        3.000e+00, 0.000e+00, 4.000e+00, 1.000e+01, 3.000e+00, 6.000e+00,\n",
       "        5.000e+00, 4.000e+00, 2.000e+00, 3.000e+00, 4.000e+00, 6.000e+00,\n",
       "        1.500e+01, 1.100e+01, 1.100e+01, 6.000e+00, 1.400e+01, 7.000e+00,\n",
       "        1.000e+01, 9.000e+00, 1.200e+01, 1.400e+01, 1.400e+01, 1.200e+01,\n",
       "        1.500e+01, 1.500e+01, 1.300e+01, 1.700e+01, 1.400e+01, 1.700e+01,\n",
       "        1.900e+01, 1.800e+01, 1.900e+01, 1.000e+01, 2.100e+01, 2.100e+01,\n",
       "        2.000e+01, 2.200e+01, 2.800e+01, 1.800e+01, 2.600e+01, 2.300e+01,\n",
       "        2.700e+01, 2.600e+01, 2.400e+01, 3.500e+01, 3.500e+01, 3.400e+01,\n",
       "        2.100e+01, 3.300e+01, 4.200e+01, 3.900e+01, 2.100e+01, 3.000e+01,\n",
       "        4.300e+01, 4.500e+01, 4.800e+01, 4.700e+01, 5.000e+01, 5.300e+01,\n",
       "        5.700e+01, 4.600e+01, 5.900e+01, 5.400e+01, 6.700e+01, 6.400e+01,\n",
       "        8.000e+01, 8.400e+01, 8.400e+01, 1.010e+02, 7.900e+01, 8.500e+01,\n",
       "        9.100e+01, 9.500e+01, 9.600e+01, 1.120e+02, 1.230e+02, 1.080e+02,\n",
       "        1.230e+02, 1.070e+02, 9.800e+01, 1.020e+02, 1.450e+02, 1.500e+02,\n",
       "        1.510e+02, 1.750e+02, 1.500e+02, 1.860e+02, 1.740e+02, 1.840e+02,\n",
       "        1.920e+02, 1.940e+02, 2.080e+02, 2.300e+02, 2.330e+02, 2.780e+02,\n",
       "        2.510e+02, 2.630e+02, 2.600e+02, 2.800e+02, 2.890e+02, 3.230e+02,\n",
       "        3.400e+02, 3.600e+02, 3.320e+02, 3.910e+02, 4.250e+02, 3.830e+02,\n",
       "        4.030e+02, 4.410e+02, 4.460e+02, 4.970e+02, 4.860e+02, 5.540e+02,\n",
       "        6.000e+02, 5.900e+02, 6.290e+02, 7.420e+02, 8.320e+02, 9.250e+02,\n",
       "        9.840e+02, 1.169e+03, 1.325e+03, 1.722e+03, 2.214e+03, 3.086e+03,\n",
       "        4.188e+03, 3.462e+03]),\n",
       " array([0.26038742, 0.26408494, 0.26778246, 0.27147998, 0.2751775 ,\n",
       "        0.27887501, 0.28257253, 0.28627005, 0.28996757, 0.29366509,\n",
       "        0.29736261, 0.30106013, 0.30475765, 0.30845516, 0.31215268,\n",
       "        0.3158502 , 0.31954772, 0.32324524, 0.32694276, 0.33064028,\n",
       "        0.33433779, 0.33803531, 0.34173283, 0.34543035, 0.34912787,\n",
       "        0.35282539, 0.35652291, 0.36022043, 0.36391794, 0.36761546,\n",
       "        0.37131298, 0.3750105 , 0.37870802, 0.38240554, 0.38610306,\n",
       "        0.38980058, 0.39349809, 0.39719561, 0.40089313, 0.40459065,\n",
       "        0.40828817, 0.41198569, 0.41568321, 0.41938073, 0.42307824,\n",
       "        0.42677576, 0.43047328, 0.4341708 , 0.43786832, 0.44156584,\n",
       "        0.44526336, 0.44896087, 0.45265839, 0.45635591, 0.46005343,\n",
       "        0.46375095, 0.46744847, 0.47114599, 0.47484351, 0.47854102,\n",
       "        0.48223854, 0.48593606, 0.48963358, 0.4933311 , 0.49702862,\n",
       "        0.50072614, 0.50442366, 0.50812117, 0.51181869, 0.51551621,\n",
       "        0.51921373, 0.52291125, 0.52660877, 0.53030629, 0.5340038 ,\n",
       "        0.53770132, 0.54139884, 0.54509636, 0.54879388, 0.5524914 ,\n",
       "        0.55618892, 0.55988644, 0.56358395, 0.56728147, 0.57097899,\n",
       "        0.57467651, 0.57837403, 0.58207155, 0.58576907, 0.58946659,\n",
       "        0.5931641 , 0.59686162, 0.60055914, 0.60425666, 0.60795418,\n",
       "        0.6116517 , 0.61534922, 0.61904674, 0.62274425, 0.62644177,\n",
       "        0.63013929, 0.63383681, 0.63753433, 0.64123185, 0.64492937,\n",
       "        0.64862688, 0.6523244 , 0.65602192, 0.65971944, 0.66341696,\n",
       "        0.66711448, 0.670812  , 0.67450952, 0.67820703, 0.68190455,\n",
       "        0.68560207, 0.68929959, 0.69299711, 0.69669463, 0.70039215,\n",
       "        0.70408967, 0.70778718, 0.7114847 , 0.71518222, 0.71887974,\n",
       "        0.72257726, 0.72627478, 0.7299723 , 0.73366982, 0.73736733,\n",
       "        0.74106485, 0.74476237, 0.74845989, 0.75215741, 0.75585493,\n",
       "        0.75955245, 0.76324996, 0.76694748, 0.770645  , 0.77434252,\n",
       "        0.77804004, 0.78173756, 0.78543508, 0.7891326 , 0.79283011,\n",
       "        0.79652763, 0.80022515, 0.80392267, 0.80762019, 0.81131771,\n",
       "        0.81501523, 0.81871275, 0.82241026, 0.82610778, 0.8298053 ,\n",
       "        0.83350282, 0.83720034, 0.84089786, 0.84459538, 0.84829289,\n",
       "        0.85199041, 0.85568793, 0.85938545, 0.86308297, 0.86678049,\n",
       "        0.87047801, 0.87417553, 0.87787304, 0.88157056, 0.88526808,\n",
       "        0.8889656 , 0.89266312, 0.89636064, 0.90005816, 0.90375568,\n",
       "        0.90745319, 0.91115071, 0.91484823, 0.91854575, 0.92224327,\n",
       "        0.92594079, 0.92963831, 0.93333583, 0.93703334, 0.94073086,\n",
       "        0.94442838, 0.9481259 , 0.95182342, 0.95552094, 0.95921846,\n",
       "        0.96291597, 0.96661349, 0.97031101, 0.97400853, 0.97770605,\n",
       "        0.98140357, 0.98510109, 0.98879861, 0.99249612, 0.99619364,\n",
       "        0.99989116]),\n",
       " <BarContainer object of 200 artists>)"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(out.prob, bins=200)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "id": "b4ecb367-6e3b-41cc-af1a-cebb252bfeb6",
   "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>molecule</th>\n",
       "      <th>species</th>\n",
       "      <th>model</th>\n",
       "      <th>prob</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>COc1cccc2c(=O)c3c(O)cc(O)c(OC)c3oc12</td>\n",
       "      <td>Frasera speciosa</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.994434</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>C=CC(C)(C)OCC1NC(=O)C2CCCN2C(=O)C(C(C)C)NC(=O)...</td>\n",
       "      <td>Lissoclinum patella</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.994601</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>C=CCc1cc(C(C)Cc2ccc(OC)c(OC)c2)c(OC)cc1O</td>\n",
       "      <td>Aniba lancifolia</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.975056</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>C=C1C2C3C(=O)OCC3C1(C)CCC2C(C)(C)O</td>\n",
       "      <td>Pyrenophora dematioidea</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.998083</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>COc1ccc(CC2COC(=O)C2C(O)c2ccc3c(c2)OCO3)cc1OC</td>\n",
       "      <td>Micranthemum umbrosum</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.954027</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34979</th>\n",
       "      <td>CC(=O)OC1CC(OC(C)=O)C2(C)C(OC(C)=O)C=CC(CCl)=C...</td>\n",
       "      <td>Junceella fragilis</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.997826</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34980</th>\n",
       "      <td>C=C1CCC2C(C)(C(=O)O)CCCC2(C)C1CCCC(=O)CO</td>\n",
       "      <td>Pinus armandii</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.975280</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34981</th>\n",
       "      <td>CCCCCCCCCCCCCC=CC(=O)N(O)CCCCCNC(=O)CCC(=O)N(O...</td>\n",
       "      <td>Nocardia</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.991955</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34982</th>\n",
       "      <td>Cc1c(O)c(C(=O)O)c(O)c2c1C(C)C(C)OC2</td>\n",
       "      <td>Penicillium</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.989203</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34983</th>\n",
       "      <td>CC1(C)C(=O)C=CC2(C)C1=C(O)C(=O)C1(C)C2CCC2(C)C...</td>\n",
       "      <td>Walsura yunnanensis</td>\n",
       "      <td>both_known</td>\n",
       "      <td>0.993248</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>22188 rows × 4 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                molecule  \\\n",
       "2                   COc1cccc2c(=O)c3c(O)cc(O)c(OC)c3oc12   \n",
       "3      C=CC(C)(C)OCC1NC(=O)C2CCCN2C(=O)C(C(C)C)NC(=O)...   \n",
       "4               C=CCc1cc(C(C)Cc2ccc(OC)c(OC)c2)c(OC)cc1O   \n",
       "5                     C=C1C2C3C(=O)OCC3C1(C)CCC2C(C)(C)O   \n",
       "6          COc1ccc(CC2COC(=O)C2C(O)c2ccc3c(c2)OCO3)cc1OC   \n",
       "...                                                  ...   \n",
       "34979  CC(=O)OC1CC(OC(C)=O)C2(C)C(OC(C)=O)C=CC(CCl)=C...   \n",
       "34980           C=C1CCC2C(C)(C(=O)O)CCCC2(C)C1CCCC(=O)CO   \n",
       "34981  CCCCCCCCCCCCCC=CC(=O)N(O)CCCCCNC(=O)CCC(=O)N(O...   \n",
       "34982                Cc1c(O)c(C(=O)O)c(O)c2c1C(C)C(C)OC2   \n",
       "34983  CC1(C)C(=O)C=CC2(C)C1=C(O)C(=O)C1(C)C2CCC2(C)C...   \n",
       "\n",
       "                       species       model      prob  \n",
       "2             Frasera speciosa  both_known  0.994434  \n",
       "3          Lissoclinum patella  both_known  0.994601  \n",
       "4             Aniba lancifolia  both_known  0.975056  \n",
       "5      Pyrenophora dematioidea  both_known  0.998083  \n",
       "6        Micranthemum umbrosum  both_known  0.954027  \n",
       "...                        ...         ...       ...  \n",
       "34979       Junceella fragilis  both_known  0.997826  \n",
       "34980           Pinus armandii  both_known  0.975280  \n",
       "34981                 Nocardia  both_known  0.991955  \n",
       "34982              Penicillium  both_known  0.989203  \n",
       "34983      Walsura yunnanensis  both_known  0.993248  \n",
       "\n",
       "[22188 rows x 4 columns]"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out[out.prob>0.95]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3e8dc2b9",
   "metadata": {},
   "source": [
    "## Discussion\n",
    "\n",
    "It seems like the model predicts well new molecules on known species but if new species are added, it is not able to predict which molecules will be present. \n",
    "\n",
    "This seems to be due to the fact that the model has been trained to predict on edges `present_in` in a \"directed\" way. So if the model has a new molecule, it takes its embeddings and is able to predict weather that molecule will be present or not.\n",
    "\n",
    "I will try to create an other model that predicts the edges on `has` type. I expect it to have the opposite result. "
   ]
  }
 ],
 "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": "3.11.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}