Fig. 4

LuCIS summarizes the severity of molecular abnormalities in individual lupus patients

Logistic regression with ridge penalization was employed to classify endotype A (Z1), the “least abnormal,” and endotype G/H (Z6), the “most abnormal,” from GSE88884 ILL-1 & ILL-2 which produced coefficients (A) that can be used to calculate LuCIS. B The mean + SEM (top) and distribution (bottom) of LuCIS calculated for the eight endotypes in all 17 datasets using the 26 features shown in A and the imputed values for the six features not measured on all platforms (IG chains, TCRA, TCRAJ, TCRB, TCRD, Treg) as described in the Supplemental Methods. Statistical differences between mean LuCIS of the endotypes were evaluated with the Kruskal–Wallis test and Dunn’s multiple comparisons. C Comparison of mean LuCIS between non-lupus healthy controls and the least “abnormal” lupus endotype in five independent datasets for which adequate controls were available. Missing values for GSVA modules not measured (IG chains, TCRAJ, TCRB, TCRD) in GSE65391 were imputed as described in the Methods. All plots were generated, and statistics were computed, in GraphPad Prism v. 9.4.0 (673)