Skip to main content
. 2019 Nov 7;9:277. doi: 10.1038/s41398-019-0623-2

Table 2.

Features included in the two prediction models

Feature Model Selection fraction Average coefficient
Proteomic
 AACT_ADLSGITGAR 2 1.00 0.122
 APOE_ALMDETMK 2 0.99 −0.195
 APOH_EHSSLAFWK 2 1.00 0.08
 FETUA_HTLNQIDEVK 2 0.97 0.082
 HBA_MFLSFPTTK 2 1.00 0.231
 PHLD_NQVVIAAGR 2 1.00 0.286
Sociodemographic
 BMI 2 1.00 0.291
 Childhood trauma 2 1.00 0.115
 Education; intermediate 2 0.93 0.065
 Education; high 2 0.93 −0.055
Clinical
 Sadness; mild 2 1.00 −0.681
 Sadness; moderate 2 1.00 0.819
 Sadness; severe 2 1.00 0.369
 Fatigue; mild 2 1.00 −0.124
 Fatigue; moderate 2 1.00 0.339
 Fatigue; severe 2 1.00 0.085
 Leaden paralysis; mild 2 1.00 −0.145
 Leaden paralysis; moderate 2 1.00 0.219
 Leaden paralysis; severe 2 1.00 0.272
 IDS30 total score 1 1.00 0.346

IDS inventory of depressive symptomatology, BMI body mass index, AACT alpha-1-antichymotrypsin, APOE apolipoprotein E, APOH apolipoprotein H, FETUA fetuin-A, HBA haemoglobin subunit alpha, PHLD glycoprotein phospholipase D

Model 1 (one feature) was based on the dominant unique model in Analysis 1 (model selection including IDS30 total score), and Model 2 (12 features) was developed by implementing feature extraction and model averaging in Analysis 2 (model selection excluding IDS30 total score) in the absence of a dominant unique model. The selection fraction and the average coefficient of the features are shown. Proteomic features are represented in a Protein_Peptide format. Categorical features (education, sadness, fatigue and leaden paralysis) are represented as sets of dummy variables