Pandas t 檢驗使用行作為 arrays

Question

我需要找到一種方法來計算兩組數據的 p 值，將一個 DataFrame 中的每一行與另一個 DataFrame 中的相應行進行比較。例如，array1 將是第 300 行中的五個項目（不包括 stdev 和 Ctrl平均值），對於第 300 行中的五個項目的 array2 也是如此。

df1:

       Pep Ctrl 1  Pep Ctrl 2   Pep Ctrl 3  Pep Ctrl 4  Pep Ctrl 5         stdev  Ctrl average
300    47591000.0         NaN   49576000.0  41288000.0  61727000.0  8.551730e+06  4.174675e+07
301     4305900.0   2670800.0          NaN         NaN   7338400.0  2.368407e+06  4.170877e+06
302    11466000.0   3799400.0          NaN  18552000.0  31661000.0  1.184124e+07  1.546393e+07
303    11255000.0   5402300.0   18337000.0  19706000.0  40286000.0  1.321849e+07  1.803413e+07

df2:

      MCI 1 vs Ctrl normalized  MCI 2 vs Ctrl normalized  MCI 3 vs Ctrl normalized  MCI 4 vs Ctrl normalized  MCI 5 vs Ctrl normalized         stdev
300               1.054045e+08              4.980206e+07              4.764870e+07              1.834201e+07              2.994124e+07  3.346473e+07
301               1.019931e+07              3.309509e+06              6.595145e+06              1.089385e+07                       NaN  3.508776e+06
302               3.288333e+07              6.953062e+06              1.430190e+07              4.988915e+06              2.310888e+07  1.162495e+07
303               3.332308e+07              1.682790e+07              2.951138e+07              9.474570e+06              2.965893e+07  1.014219e+07

我需要做一個等方差的雙尾 t 檢驗，然后將其添加為最后一列。 或者，如果 SciPy 可以選擇僅輸入項目數、標准差和平均值，這也可以。

這是我試過的：

group1 = [df1['Pep Ctrl 1'],df1['Pep Ctrl 2'],df1['Pep Ctrl 3'],df1['Pep Ctrl 4'],df1['Pep Ctrl 5']]
group2 = [df2['MCI 1 vs Ctrl normalized'], df2['MCI 2 vs Ctrl normalized'], df2['MCI 3 vs Ctrl normalized'], df2['MCI 4 vs Ctrl normalized'], df2['MCI 5 vs Ctrl normalized']]
ttest = stats.ttest_ind(a=group1,b=group2,axis = 1, equal_var = True)

任何幫助，將不勝感激。

df1構造函數：

{'Pep Ctrl 1': [47591000.0, 4305900.0, 11466000.0, 11255000.0],
 'Pep Ctrl 2': [nan, 2670800.0, 3799400.0, 5402300.0],
 'Pep Ctrl 3': [49576000.0, nan, nan, 18337000.0],
 'Pep Ctrl 4': [41288000.0, nan, 18552000.0, 19706000.0],
 'Pep Ctrl 5': [61727000.0, 7338400.0, 31661000.0, 40286000.0],
 'stdev': [8551730.0, 2368407.0, 11841240.0, 13218490.0],
 'Ctrl average': [41746750.0, 4170877.0, 15463930.0, 18034130.0]}

df2構造函數：

{'MCI 1 vs Ctrl normalized': [105404500.0, 10199310.0, 32883330.0, 33323080.0],
 'MCI 2 vs Ctrl normalized': [49802060.0, 3309509.0, 6953062.0, 16827900.0],
 'MCI 3 vs Ctrl normalized': [47648700.0, 6595145.0, 14301900.0, 29511380.0],
 'MCI 4 vs Ctrl normalized': [18342010.0, 10893850.0, 4988915.0, 9474570.0],
 'MCI 5 vs Ctrl normalized': [29941240.0, nan, 23108880.0, 29658930.0],
 'stdev': [33464730.0, 3508776.0, 11624950.0, 10142190.0]}

Answer 1

您可以使用iterrows迭代df1並將每一行與df2中具有相同索引的相應行進行比較：

from scipy import stats
df2_cols = df2.columns.drop('stdev')
out = [stats.ttest_ind(df2.loc[i, df2_cols], row, equal_var=True, nan_policy='omit') 
       for i, row in df1.drop(columns=['stdev','Ctrl average']).iterrows()]

Output：

[Ttest_indResult(statistic=0.010483243999151896, pvalue=0.9919282503324176), 
 Ttest_indResult(statistic=1.2563264347346306, pvalue=0.26449954642964396), 
 Ttest_indResult(statistic=0.009874028613226149, pvalue=0.9923973079846519), 
 Ttest_indResult(statistic=0.6390907092148139, pvalue=0.5406265164807074)]