Datasets
Models
Downloads: 1
460 lines (459 with data), 14.8 kB
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h1 align=\"center\">Machine learning-based prediction of early recurrence in glioblastoma patients: a glance towards precision medicine <br><br>[Random Forest]</h1>"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h2>[1] Library</h2>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# OS library\n",
"import os\n",
"import sys\n",
"import argparse\n",
"import itertools\n",
"import random\n",
"\n",
"# Analysis\n",
"import numpy as np\n",
"import pandas as pd\n",
"import seaborn as sns\n",
"import matplotlib.pyplot as plt\n",
"\n",
"# Sklearn\n",
"from boruta import BorutaPy\n",
"from sklearn.preprocessing import LabelEncoder\n",
"from sklearn.model_selection import train_test_split\n",
"from sklearn.ensemble import RandomForestClassifier\n",
"from sklearn.metrics import confusion_matrix, f1_score, recall_score, classification_report, accuracy_score, auc, roc_curve\n",
"from sklearn.model_selection import RandomizedSearchCV\n",
"\n",
"import scikitplot as skplt\n",
"from imblearn.over_sampling import RandomOverSampler, SMOTENC, SMOTE"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h2>[2] Exploratory data analysis and Data Preprocessing</h2>"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4>[-] Load the database</h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"file = os.path.join(sys.path[0], \"db.xlsx\")\n",
"db = pd.read_excel(file)\n",
"\n",
"print(\"N° of patients: {}\".format(len(db)))\n",
"print(\"N° of columns: {}\".format(db.shape[1]))\n",
"db.head()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4>[-] Drop unwanted columns + create <i>'results'</i> column</h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"df = db.drop(['Name_Surname','...'], axis = 'columns')\n",
"\n",
"print(\"Effective features to consider: {} \".format(len(df.columns)-1))\n",
"print(\"Creating 'result' column...\")\n",
"\n",
"# 0 = No relapse\n",
"df.loc[df['PFS'] > 6, 'result'] = 0\n",
"\n",
"# 1 = Early relapse (within 6 months)\n",
"df.loc[df['PFS'] <= 6, 'result'] = 1\n",
"\n",
"df.head()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4>[-] Check for class imbalance in the <i>'results'</i> column </h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"print(\"PFS Overview\")\n",
"print(df.result.value_counts())\n",
"\n",
"df.result.value_counts().plot(kind='pie', autopct='%1.0f%%', colors=['skyblue', 'orange'], explode=(0.05, 0.05))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4>[-] Label encoding of the categorical variables </h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"df['sex'] =df['sex'].astype('category')\n",
"df['ceus'] =df['ceus'].astype('category')\n",
"df['ala'] =df['ala'].astype('category')\n",
"\n",
"#df['Ki67'] =df['Ki67'].astype(int)\n",
"df['MGMT'] =df['MGMT'].astype('category')\n",
"df['IDH1'] =df['IDH1'].astype('category')\n",
"\n",
"df['side'] =df['side'].astype('category')\n",
"df['ependima'] =df['ependima'].astype('category')\n",
"df['cc'] =df['cc'].astype('category')\n",
"df['necrotico_cistico'] =df['necrotico_cistico'].astype('category')\n",
"df['shift'] =df['shift'].astype('category')\n",
"\n",
"## VARIABLE TO ONE-HOT-ENCODE\n",
"df['localization'] =df['localization'].astype(int)\n",
"df['clinica_esordio'] =df['clinica_esordio'].astype(int)\n",
"df['immediate_p_o'] =df['immediate_p_o'].astype(int)\n",
"df['onco_Protocol'] =df['onco_Protocol'].astype(int)\n",
"\n",
"df['result'] =df['result'].astype(int)\n",
"\n",
"dummy_v = ['localization', 'clinica_esordio', 'onco_Protocol', 'immediate_p_o']\n",
"\n",
"df = pd.get_dummies(df, columns = dummy_v, prefix = dummy_v)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4>[-] Evaluate variables' correlation with <u>'PFS'</u> columns </h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"corr = df.corr()\n",
"ax = sns.heatmap(\n",
" corr, \n",
" vmin=-1, vmax=1, center=0,\n",
" cmap=sns.diverging_palette(20, 220, n=200),\n",
" square=True\n",
")\n",
"ax.set_xticklabels(\n",
" ax.get_xticklabels(),\n",
" rotation=60,\n",
" horizontalalignment='right'\n",
");"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h2>[3] Prediction Models</h2>"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4> [-] Training and testing set splitting</h4>"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"collapsed": true
},
"outputs": [
{
"ename": "NameError",
"evalue": "name 'df' is not defined",
"output_type": "error",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)",
"\u001b[0;32m<ipython-input-5-48cdcc32916c>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtarget\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mdf\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'result'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m 2\u001b[0m \u001b[0minputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mdf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdrop\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'result'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'PFS'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m'columns'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
"\u001b[0;31mNameError\u001b[0m: name 'df' is not defined"
]
}
],
"source": [
"target = df['result']\n",
"inputs = df.drop(['result', 'PFS'], axis = 'columns')\n",
"x_train, x_test, y_train, y_test = train_test_split(inputs['...'],target,test_size=0.20, random_state=10)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4> [-] BORUTA Features Selection</h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"x = x_train.values\n",
"y = y_train.values\n",
"y = y.ravel()\n",
"\n",
"rf_boruta = RandomForestClassifier(n_jobs=-1, class_weight={0:1, 1:3}, max_depth=3)\n",
"feat_selector = BorutaPy(rf_boruta, n_estimators='auto', verbose=0, random_state=42, perc='...')\n",
"feat_selector.fit(x, y)\n",
"\n",
"cols = inputs.columns[feat_selector.support_]\n",
"print(\"N° of selected features: {}\".format(len(cols)))\n",
"print(cols) "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4> [-] Random Grid Search Hyperparameter tuning</h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# The function to measure the quality of a split\n",
"criterion = ['gini', 'entropy']\n",
"\n",
"# Number of trees in random forest\n",
"n_estimators = [int(x) for x in np.linspace(start = 20, stop = 50, num = 5)]\n",
"\n",
"# Number of features to consider at every split\n",
"max_features = ['auto', 'sqrt']\n",
"\n",
"# Maximum number of levels in tree\n",
"max_depth = [int(x) for x in np.linspace(14, 30, num = 2)]\n",
"max_depth.append(None)\n",
"\n",
"# Minimum number of samples required to split a node\n",
"min_samples_split = [ 2, 3, 4, 5, 8]\n",
"\n",
"# Minimum number of samples required at each leaf node\n",
"min_samples_leaf = [1, 2, 3, 4, 5, 6]\n",
"\n",
"max_leaf_nodes = [None, 2, 3, 4, 5, 6]\n",
"# Method of selecting samples for training each tree\n",
"bootstrap = [True, False]\n",
"\n",
"random_grid = {'criterion': criterion,\n",
" 'n_estimators': n_estimators,\n",
" 'max_features': max_features,\n",
" 'max_depth': max_depth,\n",
" 'min_samples_split': min_samples_split,\n",
" 'min_samples_leaf': min_samples_leaf,\n",
" 'max_leaf_nodes': max_leaf_nodes,\n",
" 'bootstrap':bootstrap\n",
" }\n",
"\n",
"# First create the base model to tune\n",
"rf = RandomForestClassifier(random_state=42,\n",
" n_jobs = -1, \n",
" class_weight=class_weight)\n",
"\n",
"# Random search of parameters, using 5 fold cross validation, different combinations, and use all available cores\n",
"rf_random = RandomizedSearchCV(estimator = rf, \n",
" param_distributions = random_grid, \n",
" n_iter = 500, \n",
" cv = 5)\n",
"# Fit the random search model\n",
"rf_random.fit(x_train, y_train)\n",
"rf_random.best_params_"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4>[-] SMOTE-NC</h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"smote_nc = SMOTENC(categorical_features=[3,4,10,11], k_neighbors= 3, random_state=42)\n",
"x_smote_train, y_smote_train = smote_nc.fit_resample(x_train, y_train)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4>[-] Random Forest Model</h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"rm_smote = RandomForestClassifier(random_state = 42,\n",
" criterion= '...',\n",
" n_estimators = '...',\n",
" min_samples_split = '...',\n",
" min_samples_leaf = '...',\n",
" max_leaf_nodes = '...',\n",
" max_features = '...',\n",
" max_depth = '...',\n",
" bootstrap = '...')\n",
"\n",
"rm_smote.fit(x_smote_train, y_smote_train)\n",
"print(\"Trained \\n\")\n",
"\n",
"score_rf_smote = rm_smote.score(x_test, y_test)\n",
"print(\"Random Forest accuracy: \", round(score_rf_smote*100,2), \"% \\n\")\n",
"\n",
"y_smote_predicted = rm_smote.predict(x_test)\n",
"cm_rf_smote = confusion_matrix(y_test, y_smote_predicted)\n",
"print(cm_rf_smote, \"\\n\")\n",
"\n",
"false_positive_rate, true_positive_rate, thresholds = roc_curve(y_test, y_smote_predicted)\n",
"roc_auc = auc(false_positive_rate, true_positive_rate)\n",
"\n",
"print('1. The F-1 Score of the model {} \\n '.format(round(f1_score(y_test, y_smote_predicted, average = 'macro'), 2)))\n",
"print('2. The Recall Score of the model {} \\n '.format(round(recall_score(y_test, y_smote_predicted, average = 'macro'), 2)))\n",
"print('3. Classification report \\n {} \\n'.format(classification_report(y_test, y_smote_predicted)))\n",
"print('3. AUC: \\n {} \\n'.format(roc_auc))\n",
"\n",
"tn, fp, fn, tp = cm_rf_smote.ravel()\n",
"\n",
"# Sensitivity, hit rate, Recall, or true positive rate\n",
"tpr = tp/(tp+fn)\n",
"print(\"Sensitivity (TPR): {}\".format(tpr))\n",
"\n",
"# Specificity or true negative rate\n",
"tnr = tn/(tn+fp)\n",
"print(\"Specificity (TNR): {}\".format(tnr))\n",
"\n",
"# Precision or positive predictive value\n",
"ppv = tp/(tp+fp)\n",
"print(\"Precision (PPV): {}\".format(ppv))\n",
"\n",
"# Negative predictive value\n",
"npv = tn/(tn+fn)\n",
"print(\"Negative Predictive Value (NPV): {}\".format(npv))\n",
"\n",
"# False positive rate\n",
"fpr = fp / (fp + tn)\n",
"print(\"False Positive Rate (FPV): {}\".format(fpr))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<h4> [-] Features Importance Plot </h4>"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"features = x_train.columns.values\n",
"\n",
"features[0] = 'Age'\n",
"features[6] = 'Tumor volume T1'\n",
"features[7] = 'Tumor and edema volume T2'\n",
"features[8] = 'Residual tumor'\n",
"features[9] = 'Pre-operative KPS'\n",
"features[10] = 'Post-operative KPS'\n",
"features[11] = 'Onset neurological symptoms = 1'\n",
"features[12] = 'Oncological protocol = 0'\n",
"features[13] = 'Oncological protocol = 1'\n",
"features[14] = 'Oncological protocol = 2'\n",
"\n",
"indices = np.argsort(importances)\n",
"\n",
"plt.title('Random Forest Classifier Features Importance')\n",
"plt.barh(range(len(indices)), importances[indices], color='g', align='center')\n",
"plt.yticks(range(len(indices)), [features[i] for i in indices])\n",
"plt.xlabel('Relative Importance')\n",
"\n",
"plt.savefig(\"RF Features importance.jpg\", dpi = 400, facecolor='w', edgecolor='w',\n",
" orientation='landscape', papertype=None, format=None,\n",
" transparent=False, bbox_inches='tight', pad_inches=0.3,\n",
" frameon=None)\n",
"\n",
"plt.show()"
]
}
],
"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": "3.7.4"
}
},
"nbformat": 4,
"nbformat_minor": 2
}