Non-invasive biomarkers for early diagnosis of pancreatic cancer risk: metabolite genomewide association study based on the KCPS-II cohort

Table 2 Identification of meaningful metabolites using XGBoost

Putative identification	HMBD ID	m/z	Formula	Feature importance	Pancreatic cancer incidence/control
Eicosa-11,14,17-trienoic acid	HMDB0244373	306.2560	C₂₀H₃₄O₂	6.0	1.826
Kynurenic acid	HMDB0000715	189.0429	C₁₀H₇NO₃	6.0	1.069
γ-Glutamyl tyrosine	HMDB0011741	310.1166	C₁₄H₁₈N₂O₆	5.0	1.230
N(6)-Methyllysine	HMDB0002038	160.1214	C₇H₁₆N₂O₂	5.0	0.875
LysoPE(18:0/0:0)	HMDB0011130	481.3170	C₂₃H₄₈NO₇P	5.0	1.040
Trans-3'-hydroxy cotinine	HMDB0304504	192.0901	C₁₀H₁₂N₂O₂	4.0	1.130
Palmitic amide	HMDB0012273	255.2563	C₁₆H₃₃NO	4.0	0.915
L-Leucine	HMDB0000687	131.0949	C₆H₁₃O₂	4.0	1.144
Adipic acid	HMDB0000448	146.0581	C₆H₁₀O₄	4.0	0.795
9-Decenoylcarnitine	HMDB0013205	313.2254	C₁₇H₃₁NO₄	4.0	0.794
5α-Pregnane-3,20-dione	HMDB0003759	316.2398	C₂₁H₃₂O₂	4.0	0.845

Feature Importance values > 4.0 are listed in Table 2. Feature Importance value was obtained from the XGBoost model of the training set (n = 209) [accuracy, 0.952; precision, 0.985; AUC 0.998], selecting discriminant metabolites related to pancreatic cancer incidence. The pancreatic cancer incidence/Control value was calculated using the relative abundance of each metabolite

ISSN: 1479-5876