Page 62 - 南京医科大学自然版

第44卷第12期
·1676 · 南 京 医 科 大 学 学 报 2024年12月

A B C
424241414137353324181394 0 0 35 41 40 42
n=18（0.939）。
20
2.0 10 0.90
Binomial deviance 1.5 Coefficients -10 0 10×CV accuracy 0.85

1.0
-20 0.80
-10 -8 -6 -4 -2 0 50 100 150 200 5 10 15
Lg λ L1 Norm Number of features

D E F
0.5
0.20 ST8SIA1
CHSY1
0.4
ST8SIA4
10×CV error 0.15 Error 0.3 B3GNT8
UGCG
PIGH
0.2
ALG2
0.10 B3GNT2
0.1 HAS3
SLC35A1
n=18（0.061）。 0 PIGP
MGAT4C
5 10 15 0 100 200 300 400 500 ST3GAL1
Number of features Trees ST6GALNAC6
MGAT4A
ST8SIA3
G H DPM1
LASSO SVM 1.0 ST6GALNAC3
GALNT1
0.8 SLC35B2
B3GNT3
10 6 1 GALNT14
Sensitivity 0.6 CHST9
3 ST6GAL1
5 0 0.4 ST6GALNAC5
ST8SIA1，AUC=0.865 B3GALT5
0.2
2 CHSY1，AUC=0.774 CHPF2
ST6GALNAC1
0 PIGH，AUC=0.782 MGAT5
GALNT13
00 0.2 0.4 0.6 0.8 1.0
1-Specificity
RF 0 2 4 6 8
Mean decrease Gini
I
1.0
0.8
Sensitivity 0.6 AUC：0.838

0.4
95%CI：0.747-0.917
0.2
0
00 0.2 0.4 0.6 0.8 1.0
1-Specificity
A：Selection of tuning parameter lambda in LASSO regression analysis using ten⁃fold cross⁃validation. B：LASSO coefficient profiles of diagnostic
genes. C：Line graph shows the cross⁃validated accuracy in the SVM⁃RFE model. D：Line graph shows the cross⁃validated error in the SVM⁃RFE model.
E：The effect of the number of decision trees on the error rate. F：The importance of the top 30 DEGRGs in the RF model. G：Venn diagram demonstrates
the intersection of diagnostic markers obtained from the three algorithms. H：The AUC for IgAN samples was determined using a logistic regression model.
I：ROC curves of the OFGs.
图2 糖基化相关最优特征基因的筛选和验证过程
Figure 2 The screening and validation process of the optimal feature genes related to glycosylation