Batch Correction and Comparison of Intracellular Signaling Among Multiple Rheumatic and Inflammatory Diseases

Assays that measure intracellular signaling can provide an especially mechanistic assessment of cellular states and are thereby useful for uncovering cellular physiology. Phosphoflow assays quantify responses of intracellular phosphoproteins within targeted cell populations, upon in vitro stimulation by various compounds. Having performed such assays in 11 different autoimmune and inflammatory diseases, we wished to compare the results to discover relationships within and among diseases. Batch artifacts were addressed via modeling batch means as random coefficients from a temporal stochastic process and inclusion of unaffected controls. Importantly, this batch corrective could be applied even in those extreme cases when a disease diagnosis exclusively comprised a single batch because our regression model recognized that the mean of the current batch depends upon the mean of the preceding batch and the time elapsed since the preceding assay day. Unsupervised (t-distributed stochastic neighbor embedding) and supervised (penalized supervised star plots) two-dimensional visualizations were employed for biologically meaningful summarization of artifact-corrected intracellular signaling profiles. Our results showed marked clustering of most diseases, but spreading of others, including systemic lupus erythematosus (SLE) and chronic fatigue syndrome (CFS). This approach holds promise for better understanding the physiological relationship among these diseases and for comparing longitudinally assayed datasets with unbalanced batching in general.