Instability-correlated gene expression signature and regression modeling. To identify an instability-correlated gene expression signature, instability index was modeled as a function of gene expression using the mouse Gene Expression Atlas. We calculated the correlation between instability index and expression level, and built regression models by sequentially introducing top n number of the most highly correlated probes with instability index (16 training tissues, 2 gene expression replicates) using partial least square regression (PLSR). The lowest error rate (root mean squared error of prediction, RMSEP) in leave one out cross validation (0.235) was obtained by modeling of the 150 most correlated probes. The predictive power of the model was verified by two independent test sets. Firstly, we determined instability indices of 4 additional tissues, muscle, olfactory bulb, white adipose tissue and adrenal gland (HdhQ 111/+, 5 months, n = 4-6 mice), and compared them with instability indices predicted by the regression model (blue, 2 gene expression replicates). Secondly, we predicted instability indices using independent striatum and cerebellum microarray data (GSE9025, HdhQ 111/+, 5 months, n = 1), and compared them to measured instability indices (red). RMSEP, root mean squared error of prediction.