Skip to main content
. 2021 Apr 14;8:632950. doi: 10.3389/fmolb.2021.632950

TABLE 3.

Feature inclusion frequency using LASSO based feature selection for MI (vs. UA).

Metabolites-only model Metabolites and risk factor model
Variable Frequency Coefficient Variable Frequency Coefficient
1 TG (14:0/20:3/22:2) 200 0.16 Age 200 −0.02
2 3-Hydroxy-3-methyl butyric acid 200 0.13 LVEF 200 −0.05
3 HexCer(d18:1/22:0) 200 0.08 HexCer(d18:1/22:1) 200 −0.13
4 HexCer(d18:1/22:1) 200 −0.16 3-Hydroxy-3-methyl butyric acid 200 0.19
5 PA (36:1) 200 −0.17 CerP (d18:1/20:3) 199 −0.08
6 PC(O-38:2) 200 −0.22 Cer(d18:1/22:1) 185 −0.05
7 D-Methionine 199 0.12 PC (18:2/18:2) 174 −0.02
8 Deoxycholic acid 199 −0.06 PC (16:0/20:3) 166 0.15
9 PC (18:2/18:2) 199 −0.12 HDLC 149 −0.04
10 HexCer (d18:1/26:1) 182 0.06 Hypertension 126 −0.04
11 PC (O-34:2) 182 −0.03 PC (18:2/20:4) 126 −0.04
12 PC (18:2/20:2) 182 −0.07 PC(O-38:2) 126 −0.02
13 L-Cystine 137 0.02 Deoxycholic acid 126 −0.01
14 TG (14:1/16:1/22:3) 137 −0.02 L-Cystine 101 0.01
15 3-Methylcrotonyl glycine 104 0.02 D-Methionine 101 0.02
16 MG (18:1) 104 −0.05

LASSO based feature selection was performed within a 5-fold cross-validation framework (200 iterations). Variables selected with frequency >100 times and their average coefficient (Coefficient) were indicated. MI, myocardial infarction; PA, phosphatidic acid; MG, monoglyceride, other abbreviations as in Table 4.