Table 3 Feature importances for the classifier predicting MGD level 2 from relative protein quantifications using SHAP weighted by model uncertainty.

From: Using machine learning model explanations to identify proteins related to severity of meibomian gland dysfunction

Rank

Protein (accession number), incl. BL

Protein (accession number), excl. BL

1

Tissue alpha-L-fucosidase (P04066)

Tissue alpha-L-fucosidase (P04066)

2

Mannose-1-phosphate guanyltransferase alpha (Q96IJ6)

Mannose-1-phosphate guanyltransferase alpha (Q96IJ6)

3

UDP-glucose 4-epimerase (Q14376)

UDP-glucose 4-epimerase (Q14376)

4

IGKV2-24 (A0A0C4DH68)

Protein S100-A8 (P05109)

5

Polymeric immunoglobulin receptor (P01833)

Polymeric immunoglobulin receptor (P01833)

6

Protein S100-A8 (P05109)

IGKV2-24 (A0A0C4DH68)

7

CYFIP-related Rac1 interactor B (Q9NUQ9)

SH2 domain-containing protein 4A (Q9H788)

8

60S acidic ribosomal protein P0 (P05388)

Adenylate kinase isoenzyme 1 (P00568)

9

Adenylate kinase isoenzyme 1 (P00568)

60S acidic ribosomal protein P0 (P05388)

10

SH2 domain-containing protein 4A (Q9H788)

Lactoylglutathione lyase (Q04760)

11

Lactoylglutathione lyase (Q04760)

CYFIP-related Rac1 interactor B (Q9NUQ9)

12

IGA2 (P0DOX2)

IGA2 (P0DOX2)

13

Dynactin subunit 2 (Q13561)

Lysozyme C (P61626)

14

Lysozyme C (P61626)

PRP4 (Q16378)

15

Alpha-centractin (P61163)

Serotransferrin (P02787)

  1. Bold font: Not among the original top ranked features. BL, borderline predictions.