drug-AM.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "am = pd.DataFrame.from_csv('Drug_expanded/drug-drug-AM.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "for c in am.columns:\n",
    "        if not am[c].equals(am.loc[c]):\n",
    "            print(c)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "df = pd.DataFrame.from_csv('Train_data/clustered2.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "df['Target_share'] = \"\"\n",
    "for i in range(df.shape[0]):\n",
    "    df.set_value(i, 'Target_share', am.ix[df.ix[i, 'COMPOUND_A'], df.ix[i, 'COMPOUND_B']])\n",
    "    #df['Target_share'][i] = am[df['COMPOUND_A'][i]][df['COMPOUND_B'][i]] "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['CELL_LINE', 'COMPOUND_A', 'COMPOUND_B', 'MAX_CONC_A', 'MAX_CONC_B',\n",
       "       'IC50_A', 'H_A', 'Einf_A', 'IC50_B', 'H_B', 'Einf_B', 'SYNERGY_SCORE',\n",
       "       'A_HBA', 'A_cLogP', 'A_HBD', 'A_Lipinski', 'A_MW', 'A_ALogP/XLogP',\n",
       "       'A_PSA', 'A_#RotBonds', 'A_Arom Rings', 'A_Heavy atoms', 'A_QED',\n",
       "       'B_HBA', 'B_cLogP', 'B_HBD', 'B_Lipinski', 'B_MW', 'B_ALogP/XLogP',\n",
       "       'B_PSA', 'B_#RotBonds', 'B_Arom Rings', 'B_Heavy atoms', 'B_QED',\n",
       "       'GEX_CLUSTER', 'ORIGIN_CLUSTER', 'THERAPY_CLUSTER', 'Target_share'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 129,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df.to_csv('Train_data/with_targets2.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "df = df.iloc[np.random.permutation(len(df))]\n",
    "Ypd = df['SYNERGY_SCORE']\n",
    "Xpd = pd.get_dummies(df.drop('SYNERGY_SCORE', axis=1), columns=['CELL_LINE', 'COMPOUND_A', 'COMPOUND_B'])\n",
    "\n",
    "#orig_data = pd.DataFrame.from_csv('Clean_data/train2.csv', header=0, index_col=None)\n",
    "#df = df.iloc[np.random.permutation(len(orig_data))]\n",
    "#Xpd = pd.get_dummies(df, columns=['CELL_LINE', 'COMPOUND_A', 'COMPOUND_B'])\n",
    "#Ypd = orig_data['SYNERGY_SCORE'].reindex(Xpd.index)\n",
    "\n",
    "Xvals = Xpd.values\n",
    "yvals = Ypd.values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def print_scores(model):\n",
    "    print(\"R^2: \" + str(model.score(X_test, y_test)))\n",
    "    print(\"RMSE: \" + str(scoring.rmse(y_test, model.predict(X_test))) ) # root mean squared error\n",
    "    print(\"MAE: \" + str(metrics.mean_absolute_error(y_test, model.predict(X_test)))) # mean absolute error\n",
    "    print(\"MAPE: \" + str(scoring.mape(y_test, model.predict(X_test)))) # mean absolute percent error\n",
    "    print(\"Error within 20: \" + str(scoring.error_within(0.2, y_test, model.predict(X_test)))) # error within\n",
    "    print(\"Pearson: \" + str(stats.pearsonr(y_test, model.predict(X_test)))) # pearson correlation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.ensemble import GradientBoostingRegressor\n",
    "from sklearn.svm import SVR, SVC\n",
    "from sklearn import cross_validation, metrics\n",
    "from sklearn.grid_search import GridSearchCV, RandomizedSearchCV\n",
    "import scipy.stats as stats\n",
    "import scoring"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def pearson(y, y_pred):\n",
    "    return stats.pearsonr(y, y_pred)[0]\n",
    "pearson_scorer = metrics.make_scorer(pearson)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.350100795095, 0.617507234405\n",
      "0.363878057625, 0.622800835484"
     ]
    }
   ],
   "source": [
    "for i in range(3):\n",
    "    gbr = GradientBoostingRegressor(n_estimators=1000, max_depth=5, loss='huber',\n",
    "                                max_features='log2', learning_rate=0.03)\n",
    "    pearsons = cross_validation.cross_val_score(gbr, Xvals, yvals, scoring=pearson_scorer, cv=4)\n",
    "    r2s = cross_validation.cross_val_score(gbr, Xvals, yvals, cv=4)\n",
    "\n",
    "    print(str(np.mean(r2s)) + ', ' + str(np.mean(pearsons)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = cross_validation.train_test_split(Xvals, yvals)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "GradientBoostingRegressor(alpha=0.9, init=None, learning_rate=0.03, loss='ls',\n",
       "             max_depth=7, max_features='log2', max_leaf_nodes=None,\n",
       "             min_samples_leaf=1, min_samples_split=2,\n",
       "             min_weight_fraction_leaf=0.0, n_estimators=1000,\n",
       "             random_state=None, subsample=1.0, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 126,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gbr = GradientBoostingRegressor(n_estimators=1000, max_depth=7,\n",
    "                                max_features='log2', learning_rate=0.03)\n",
    "gbr.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "R^2: 0.303345939352\n",
      "RMSE: 38.27535550342086\n",
      "MAE: 23.0766183463\n",
      "MAPE: 3.83356707243\n",
      "Error within 20: 0.489795918367\n",
      "Pearson: (0.55289669457883595, 1.7725165894405088e-44)\n"
     ]
    }
   ],
   "source": [
    "print_scores(gbr)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.352857325976, 0.595579294764\n",
      "0.233894086761, 0.536985038475\n",
      "0.264293977878, 0.542710665055\n",
      "0.29200569844, 0.593830099557\n",
      "0.336467592446, 0.595107481029\n",
      "0.454032319331, 0.689335673817\n",
      "0.397183809878, 0.630676384758\n",
      "0.385285222895, 0.623321078576\n",
      "0.399223534995, 0.645433127192\n",
      "0.32266407888, 0.606421353709\n",
      " \n",
      "Average: 0.343790764748, 0.605940019693\n"
     ]
    }
   ],
   "source": [
    "r2s = []\n",
    "pearsons = []\n",
    "for i in range(10):\n",
    "    X_train, X_test, y_train, y_test = cross_validation.train_test_split(Xvals, yvals)\n",
    "    gbr = GradientBoostingRegressor(n_estimators=1000, max_depth=7,\n",
    "                                max_features='log2', learning_rate=0.03)\n",
    "    gbr.fit(X_train, y_train)\n",
    "    r2 = gbr.score(X_test, y_test)\n",
    "    r2s.append(r2)\n",
    "    p = stats.pearsonr(y_test, gbr.predict(X_test))[0]\n",
    "    pearsons.append(p)\n",
    "    \n",
    "    print(str(r2) + \", \" + str(p))\n",
    "print(\" \")\n",
    "print(\"Average: \" + str(np.mean(r2s)) + \", \" + str(np.mean(pearsons)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 170,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "param_grid_gb = {\n",
    "    'learning_rate': [0.03, 0.05],\n",
    "    'n_estimators': [100, 200, 300],\n",
    "    'max_features': ['log2', 'sqrt', 'auto'],\n",
    "    'max_depth': [7, 8]\n",
    "}\n",
    "gbr = GradientBoostingRegressor()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 171,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'learning_rate': 0.03, 'max_depth': 8, 'max_features': 'log2', 'n_estimators': 300}\n",
      "0.320772621424\n",
      "{'learning_rate': 0.05, 'max_depth': 7, 'max_features': 'log2', 'n_estimators': 300}\n",
      "0.319446549601\n",
      "{'learning_rate': 0.05, 'max_depth': 8, 'max_features': 'log2', 'n_estimators': 200}\n",
      "0.317491377939\n"
     ]
    }
   ],
   "source": [
    "grid_search_gb = GridSearchCV(gbr, param_grid_gb, n_jobs=2)\n",
    "for i in range(3):\n",
    "    grid_search_gb.fit(Xvals, yvals)\n",
    "    print(grid_search_gb.best_params_)\n",
    "    print(grid_search_gb.best_score_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline \n",
    "plt.rcParams['figure.figsize'] = (10, 6)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAnEAAAF/CAYAAAAmb5/VAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nWBfHQwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x107200a90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = y_test\n",
    "y = gbr.predict(X_test)\n",
    "plt.scatter(x, y)\n",
    "plt.plot(x, x, 'k--')\n",
    "\n",
    "m, b = np.polyfit(x, y, 1)\n",
    "plt.plot(x, m*x + b, 'b--')\n",
    "\n",
    "plt.axis([-200, 150, -200, 150])\n",
    "plt.xlabel(\"True synergy\")\n",
    "plt.ylabel(\"Predicted synergy\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Classification"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import roc_auc_score\n",
    "from sklearn.preprocessing import StandardScaler"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "Xstand = StandardScaler().fit_transform(Xvals)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def svm_pred():\n",
    "    Yclass = Ypd > 0\n",
    "    Yclass = Yclass.astype(int)\n",
    "    Yvals = Yclass.values\n",
    "\n",
    "#     m = Ypd.shape[0]\n",
    "#     Yvals = np.zeros(m)\n",
    "#     for i in range(m):\n",
    "#         if Ypd[i] > 50:\n",
    "#             Yvals[i] = 1\n",
    "#         elif Ypd[i] < -50:\n",
    "#             Yvals[i] = -1\n",
    "#         else:\n",
    "#             Yvals[i] = 0\n",
    "    X_train, X_test, y_train, y_test = cross_validation.train_test_split(Xstand, Yvals)\n",
    "    \n",
    "    svm = SVC() #(C=10, gamma=0.001)\n",
    "    svm.fit(X_train, y_train)\n",
    "    pred = svm.predict(X_test)\n",
    "    print(\"Classification accuracy: \" + str(svm.score(X_test, y_test)))\n",
    "    print(\"AUC: \" + str(metrics.roc_auc_score(y_test, pred)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Classification accuracy: 0.667903525046\n",
      "AUC: 0.60544242389\n"
     ]
    }
   ],
   "source": [
    "svm_pred()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def grid_search_svm():\n",
    "    params = {\n",
    "        'C': [1, 5, 10],\n",
    "        'gamma': [0.005, 0.001, 0.0005]\n",
    "    }\n",
    "    svc = SVC()\n",
    "    \n",
    "    Yclass = Ypd > 0\n",
    "    Yclass = Yclass.astype(int)\n",
    "    Yvals = Yclass.values\n",
    "    \n",
    "    search = GridSearchCV(svc, params, n_jobs=2)\n",
    "    for i in range(3):\n",
    "        search.fit(Xstand, Yvals)\n",
    "        print(search.best_params_)\n",
    "        print(search.best_score_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'gamma': 0.001, 'C': 10}\n",
      "0.836118848654\n",
      "{'gamma': 0.001, 'C': 10}\n",
      "0.836118848654\n",
      "{'gamma': 0.001, 'C': 10}\n",
      "0.836118848654\n"
     ]
    }
   ],
   "source": [
    "grid_search_svm()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.4.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}