Perezet al.,Science 376 , eabf1970 (2022) 8 April 2022 7 of 13
AB
Thrombocytopenia
Renal involvement
Tup
Bup
anti-Smith
SLADAI
Low complement
Panup
Myeup
Combined
Leukopenia
Lymphopenia
Discoid rash
Malar rash
Photosensitivity
Myedown
Neurological involvement
anti-dsDNA
Serositis
Pandown
Mucosal Ulcers
Arthritis
anti-phospholipid
IFITM3 Myeup
ISG15 Panup
HLA-C Panup
MT2APanup
LY6E Panup
S100A11 Myeup
IFI27 Panup
IFI6 Panup
TYROBPMyeup
TYMPMyeup
MNDAMyeup
JUNB Bup
IFITM2 Panup
TCL1ABup
PSME2 Panup
LGALS1 Myeup
IFI44LPanup
VPREB3 Bup
IFITM1 Myeup
TIMP1 Myeup
PSMB9 Panup
LGALS2 Myeup
TNFSF10 Myeup
TNFSF13B Myeup
Tup
Bup
Panup
Myeup
Myedown
Pandown
Tup
Bup
Panup
Myeup
Myedown
Pandown
C
anti-Smith
Mucosal Ulcers Thrombocytopenia
Mean AUC = 0.70 0.09
True positive rate
1.0
1.5
0.0 Odds ratio
RNASE2PLD4
STRADBNPDC1
VPREB3RTKN2
RP11-1399P15.1LGALS3BP
IFI27
anti-dsDNA
Mean AUC = 0.66 0.00
Odds ratio
FKBP5ARRDC3
SAMD3CELA1
AK5APOBEC3A
MDS2ZDHHC1
BHLHE40TSHZ2
Mean AUC = 0.69 0.08
True positive rate
1.0
1.5
0.0
Mean AUC = 0.680.01
Odds ratio
IL1BNR4A1
G0S2FRG1B
PATL2ARHGAP24
RP5-887A10.1CA11
SDR42E2PTCH2
Odds ratio
LILRA3GPR183
VMO1KIF22
LINC00649PTCH2
LGALS9ODF3B
ICAM4PTGER4
Lorem ipsum
0.0 0.5 1.0 0.0 0.5 1.0
01 2 01 2
(^02012)
D
Odds ratio
0 1 2
Differentially
expressed in:
True positive rate
Case control OOS
ROC AUC: 0.84
1.0
0.8
0.6
0.4
0.2
0.0
0.0 0.5 1.0
VRK2
AC016629.8
TCL1A
STRADB
MS4A6A
GBP3
TNFSF13B
FOS
EIF3L
PITPNC1
DNAJB1
SDR42E2
CTD-2035E11.3
MDS2
PRMT10
E
Molecular PC1
Molecular PC2
CTL
Training Set
SLE
Molecular Clusters
Low High
Molecular Clusters
F
- 10
5
0
Low
Training Set
Low: 1.8, High: 4.0
False positive rate False positive rate False positive rate
High
SLEDAI
H
Log
T cell abundance 10
Log 10 module expression
anti-dsDNAanti-Smith
Renal involvementDiscoid rash
LymphopeniaMalar rash
PhotosensitivityLeukopenia
Low complementArthritis
Neurological involvementSerositis
Thrombocytopeniaanti-phospholipid
Mucosal Ulcers
0 2 4
Odds ratio
Test Set
given high group membershipOdds of clinical feature
Up-Pan R= -0.55
Asian R= -0.52, European R= -0.57
IFITM3 Myeup
ISG15 Panup
HLA-C Panup
MT2APanup
LY6E Panup
S100A11 Myeup
IFI27 Panup
IFI6 Panup
TYROBPMyeup
TYMPMyeup
MNDAMyeup
JUNB Bup
IFITM2 Panup
TCL1ABup
PSME2 Panup
LGALS1 Myeup
IFI44LPanup
VPREB3 Bup
IFITM1 Myeup
TIMP1 Myeup
PSMB9 Panup
LGALS2 Myeup
TNFSF10 Myeup
TNFSF13B Myeup
CTL Low High
G
Molecular PC1
Test set
Molecular PC2
SLE Low High
European
Asian
Correlation
0 1
Fig. 4. Prediction of disease status and molecular stratification of SLE.
(A) Correlation between log 10 (expression of Panup)(xaxis) and log 10 (abundance
of CD4Naïvecells) in processing batch 4 cases only. (B) Correlation matrix
between average expression of each of six gene modules and clinical features.
(CandD) Receiver operating curve for out-of-sample (OOS) prediction of case-
control status (C) and individual clinical variables (D) using a logistic regression
model trained on 302 expression features. Inset depicts the most important
molecular features inferred by the model, colored by the module to which each
feature belongs. (E) Principal components analysis of training set based on
302 expression features. Green, control; red, case. Heatmap shows the top
25 most correlated expression features to molecular principal component PC1.
Expression was binned and averaged across 24 equal steps across molecular
PC1. K-means clustering of samples based on principal components yielded
two molecular subphenotypes (Low, High). (F) Distribution of SLEDAI scores
(yaxis) for each molecular subphenotype (xaxis) in the training cases. *P< 0.05
(Wilcoxon rank-sum test). (G) Projection of OOS test set onto molecular PC1
and PC2 and colored by case-control status (left) and molecular cluster membership
(right). Heatmap shows the top 25 most correlated expression features to
molecular PC1 in the test set. (H) Odds ratio of having a clinical feature given
membership in the High molecular cluster versus the Low molecular cluster.
RESEARCH | RESEARCH ARTICLE