簡介
通過基因表達監測（DNA微陣列）對新的癌症病例進行分類，從而爲鑑定新的癌症類別和將腫瘤分配到已知類別提供了一般方法。這些數據用於對患有急性髓性白血病（AML）和急性淋巴細胞白血病（ALL）的患者進行分類。
代碼實例
導入依賴庫
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
載入數據
labels_df = pd.read_csv('actual.csv', index_col = 'patient')
test_df = pd.read_csv('data_set_ALL_AML_independent.csv')
data_df = pd.read_csv('data_set_ALL_AML_train.csv')
print('train_data: ', data_df.shape, '\n test_data: ', test_df.shape, '\n labels: ', labels_df.shape)
# labels_df.shape
data_df.head() #查看前5行（默認前5行）
清理數據
test_cols_to_drop = [c for c in test_df.columns if 'call' in c]
test_df = test_df.drop(test_cols_to_drop, axis=1)
test_df = test_df.drop(['Gene Description', 'Gene Accession Number'], axis=1 )
data_cols_to_drop = [c for c in data_df.columns if 'call' in c]
data_df = data_df.drop(data_cols_to_drop, axis=1)
data_df = data_df.drop(['Gene Description', 'Gene Accession Number'], axis=1 )
print('train_data ', data_df.shape, '\n test_data: ', test_df.shape, '\n labels: ', labels_df.shape)
data_df.head()
定義'特徵'和'樣本'
使用基因表達值來預測癌症類型。 因此，特徵是患者的基因和樣本。 使用X作爲輸入數據，其中行是樣本（患者），列是特徵（基因）。
將'ALLL'替換爲0，將'AML'替換爲1。
labels_df = labels_df.replace({'ALL':0, 'AML':1})
train_labels = labels_df[labels_df.index <= 38]
test_labels = labels_df[labels_df.index > 38]
print(train_labels.shape, test_labels.shape)
# labels_df.index
test_df = test_df.T
train_df = data_df.T
檢查空值
print('Columns containing null values in train and test data are ', data_df.isnull().values.sum(), test_df.isnull().values.sum())
聯合訓練集和測試集
full_df = train_df.append(test_df, ignore_index=True)
print(full_df.shape)
full_df.head()
標準化和處理高維度
標準化
所有變量具有非常相似的範圍的方式重新調整我們的變量（預測變量）。
高維度
只有72個樣本和7000多個變量。 意味着如果採取正確的方法，模型很可能會受到HD的影響。 一個非常常見的技巧是將數據投影到較低維度空間，然後將其用作新變量。 最常見的尺寸減小方法是PCA。
# Standardization
from sklearn import preprocessing
X_std = preprocessing.StandardScaler().fit_transform(full_df)
# Check how the standardized data look like
gene_index = 1
print('mean is : ', np.mean(X_std[:, gene_index] ) )
print('std is :', np.std(X_std[:, gene_index]))
fig= plt.figure(figsize=(10,10))
plt.hist(X_std[:, gene_index], bins=10)
plt.xlim((-4, 4))
plt.xlabel('rescaled expression', size=30)
plt.ylabel('frequency', size=30)
PCA(聚類分析)
# PCA
from sklearn.decomposition import PCA
pca = PCA(n_components=50, random_state=42)
X_pca = pca.fit_transform(X_std)
print(X_pca.shape)
cum_sum = pca.explained_variance_ratio_.cumsum()
cum_sum = cum_sum*100
fig = plt.figure(figsize=(10,10))
plt.bar(range(50), cum_sum)
plt.xlabel('PCA', size=30)
plt.ylabel('Cumulative Explained Varaince', size=30)
plt.title("Around 90% of variance is explained by the First 50 columns ", size=30)
labels = labels_df['cancer'].values
colors = np.where(labels==0, 'red', 'blue')
from mpl_toolkits.mplot3d import Axes3D
plt.clf()
fig = plt.figure(1, figsize=(15,15 ))
ax = Axes3D(fig, elev=-150, azim=110,)
ax.scatter(X_pca[:, 0], X_pca[:, 1], X_pca[:, 2], c=colors, cmap=plt.cm.Paired,linewidths=10)
ax.set_title("First three PCA directions")
ax.set_xlabel("PCA1")
ax.w_xaxis.set_ticklabels([])
ax.set_ylabel("PCA2")
ax.w_yaxis.set_ticklabels([])
ax.set_zlabel("PCA3")
ax.w_zaxis.set_ticklabels([])
plt.show()
RF分類
X = X_pca
y = labels
print(X.shape, y.shape)
劃分訓練集和測試集
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)
print(X_train.shape, y_train.shape)
import seaborn as sns
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score
rfc = RandomForestClassifier(random_state=42)
rfc.fit(X_train, y_train)
y_pred = rfc.predict(X_test)
print(accuracy_score(y_test, y_pred))
0.7083333333333334
重要特徵
labs = ['PCA'+str(i+1) for i in range(X_train.shape[1])]
importance_df = pd.DataFrame({
'feature':labs,
'importance': rfc.feature_importances_
})
importance_df_sorted = importance_df.sort_values('importance', ascending=False)
importance_df_sorted.head()
fig = plt.figure(figsize=(25,10))
sns.barplot(data=importance_df_sorted, x='feature', y='importance')
plt.xlabel('PCAs', size=30)
plt.ylabel('Feature Importance', size=30)
plt.title('RF Feature Importance', size=30)
plt.savefig('RF Feature Importance.png', dpi=300)
plt.show
Confusion Matrix（混淆矩陣）
Gradient Boost Classifier（梯度增強分類器）
from sklearn.metrics import confusion_matrix
cm = confusion_matrix(y_test, y_pred)
### Normalize cm, np.newaxis makes to devide each row by the sum
cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
print(np.newaxis)
cmap=plt.cm.Blues
plt.imshow(cm, interpolation='nearest', cmap=cmap)
print(cm)
from sklearn.ensemble import GradientBoostingClassifier
gbc = GradientBoostingClassifier(random_state=42)
gbc.fit(X_train, y_train)
y_gbc_pred = gbc.predict(X_test)
print(accuracy_score(y_test, y_gbc_pred))
0.7083333333333334
參考：
Molecular Classification of Cancer: Class Discovery and Class Prediction by Gene Expression
Science 286:531-537. (1999). Published: 1999.10.14
T.R. Golub, D.K. Slonim, P. Tamayo, C. Huard, M. Gaasenbeek, J.P. Mesirov, H. Coller, M. Loh, J.R. Downing, M.A. Caligiuri, C.D. Bloomfield, and E.S. Lander