import pandas as pd
import numpy as np
from rdkit.ML.Scoring import metrics
# the metrics is stolen from oddt
# https://github.com/oddt/oddt

读入打分结果

df = pd.read_csv('/public/gkxiao/work/demo/pde5_score.csv')

预览

df.head(5)

	score	label
0	-9.971069	active
1	-9.694971	active
2	-11.613330	active
3	-9.017618	active
4	-12.542497	active

按打分值从低到高排列，打分最好的那个排第一

df_sorted = df.sort_values(by=['score'],ascending=True)

预览

df_sorted.head(5)

	score	label
32	-16.976294	active
25	-16.287430	active
10	-16.262144	active
11	-15.785425	active
24	-15.641940	active

将打分值乘以-1，转化为正值以便进行性能计算

score = df_sorted['score']
x = np.array(score)*(-1)

将实验值active与decoy转化为1与0

y = df_sorted['label']
label_map = {'active': 1, 'decoy': 0}
labels = [label_map[y_true] for y_true in y]
y = np.array(labels)

计算AUC

auc = metrics.roc_auc(y, x, pos_label=1, ascending_score=False)
print("ROC AUC = ",auc)

ROC AUC =  0.7996552281420028

计算alpha=20.0的BEDROC

bedroc = metrics.bedroc(y, x, alpha=20.0, pos_label=1)
print("alpha=20.0 BEDROC = ",bedroc)

alpha=20.0 BEDROC =  0.4518017076489426

计算top 1%的富集因子

ef = metrics.enrichment_factor(y, x, percentage=1, pos_label=1, kind='fold')
print("EF at 1% = ",ef)

EF at 1% =  13.991228070175438

计算[0.0001,1]的半对数AUC

logauc = metrics.roc_log_auc(y, x, pos_label=1, ascending_score=False, log_min=0.001, log_max=1.0)
print("logAUC [0.1%,100%]  = ",logauc)

logAUC [0.1%,100%]  =  0.4205614548770312