Transient and etiology-related transcription regulation in cirrhosis prior to hepatocellular carcinoma occurrence

Table 3 Performance of various, unsupervised or supervised classification tools for HCC-free cirrhosis samples

Samples2	Tool1
	UHC (%)	SVM (%)	KNNC (%)
A1-A7 + A15-A24	653	100	100
V8-V14 + V25-V34	71	80	70
CL1-CL18	100	70	80
All test samples	79	83	83

¹Unsupervised hierarchical clustering (UHC) was done with 52 cirrhotic or CL samples. Supervised training/testing procedures (KNNC; SVM) were each done by first separating these 52 samples into 22 training (A1-A7, V8-V14 and CL1-CL8) and 30 test samples (A15-A24, V25-V34 and CL9-CL18).

²20 transcript levels were measured in total RNA from every sample by qRT-PCR. These transcripts had the most significant difference in levels (Bonferroni-corrected ANOVA, P < 0.05) between alcoholism vs HCV vs CLs (ACSM2, ADH4, AHNAK, APOH, ARID1A, ARL2BP, ATP5G2, CYP2E1, CYP3A4, DPF2, DPM1, GSTA 1-2, HSCARG, HSD17B6, IFI27, MT1G, ORM2, PLG, STAT3, TIMP1).

³% of properly classified test samples.