Fig. 7: Evaluation of technical sources of variation and reproducibility.

We analyzed data from on-chip readouts across all studies using a mixed model, which is a statistical model that can account for different sources of variability. By using the output from this model, we can understand better how technical factors affect the endpoints of interest and can show that even in the presence of this technical noise, we can still pick up biological effects of interest with sufficient precision. a, b The proportion (%) of total variance associated with each of the technical factors (circuit, study, laboratory) and the proportion of residual variance (that is, the variance that cannot be attributed to any of the other factors). Proportions are plotted respectively for the endpoints albumin (a) and ketone bodies (b). c, d We viewed the difference between the diseased and healthy conditions on day 13 as a benchmark biological difference of interest. Then, we used the mixed model to both estimate that difference, and also quantify how precise our estimate is. For each study, the difference and its associated standard error are shown for albumin (c) and ketone bodies (d). Note that in this case, the statistical model was applied to each study independently. Studies 1 and 2 were performed at TissUse and Study 3 at AstraZeneca.