Fig. 7: Supervised machine learning (ML) models to predict type 1 diabetes (T1D) from recursive feature elimination (RFE)-selected proteins, with and without inclusion of high-risk HLA allele dosage.
From: The inflammatory path toward type 1 diabetes begins during pregnancy
a Performance of the Olink-only model (without HLA dosage), using eXtreme Gradient Boosting (XGBoost), based on b the top 30 recursive feature elimination (RFE)-selected proteins using an 80% training and 20% testing set from in the ABIS Olink case/control cohort (n = 432; cases, n = 146; controls, n = 286). c Performance of the Olink + HLA machine learning (ML) model, based on d the top 30 RFE-selected features. e Median difference of the RFE-selected proteins in the Olink + HLA model. f Normalized protein expression (NPX) values of HLA-DRA by high-risk HLA allele dosage (0, 1, or 2 copies of DR4DQ8 or DR3DQ2). Boxplots show the median (line) and interquartile range (box, 25th–75th percentile) to demonstrate the data distribution. Whiskers extend to the most extreme values within 1.5 × interquartile range (IQR) of the lower and upper quartiles; points beyond the whiskers are plotted as outliers. AUC-ROC area under the receiver operating characteristic curve, ALDH3A1 aldehyde dehydrogenase 3 family member A1, ARHGEF12 Rho guanine nucleotide exchange factor 12, AXIN1 axis inhibition protein 1, BCR breakpoint cluster region protein, BTN3A2 butyrophilin subfamily 3 member A2, CASP2 caspase-2, CD48 CD48 molecule, CD160 CD160 molecule, CXCL14 C-X-C motif chemokine ligand 14, DNER delta and notch-like epidermal growth factor-related receptor, FXYD5 FXYD domain-containing ion transport regulator 5, GAL galanin and GMAP prepropeptide, HLA_DRA major histocompatibility complex, class II, DR alpha, IDS iduronate 2-sulfatase, IL1B interleukin-1 beta, IL13 interleukin-13, IL20RA interleukin-20 receptor subunit alpha, IL22RA1 interleukin-22 receptor subunit alpha-1, IL1RL2 interleukin-1 receptor-like 2, LAIR1 leukocyte-associated immunoglobulin-like receptor 1, NME3 NME/NM23 family member 3, OSCAR osteoclast-associated receptor, PADI1 peptidyl arginine deiminase 1, PADI2 peptidyl arginine deiminase 2, PTH1R parathyroid hormone 1 receptor, SCGB3A2 secretoglobin family 3A member 2, SERPINB8 serpin family B member 8, SIT1 signaling threshold-regulating transmembrane adapter 1, TIMP3 tissue inhibitor of metalloproteinases 3, TNFRSF14 tumor necrosis factor receptor superfamily member 14, TNFSF12 tumor necrosis factor ligand superfamily member 12, TREM2 triggering receptor expressed on myeloid cells 2.
