Fig. 5: Linear models for estimation of the Spillover Spreading Matrix (SSM).

Examples are shown for the datasets MM1 (left, a, c, e) and HS1 (right, b, d, f). a, b Regression carried out over the gated events of one single-color control of each dataset, with no well-defined positive and negative populations, with the primary and secondary channels as indicated, respectively, in the y- and x-axes. Uncompensated data points are displayed in blue and compensated ones in black. Regression from uncompensated (resp. compensated) data is displayed with dashed (resp. solid) lines, in black (resp. gray) when the regression coefficient is significant and positive (resp. non-significant or non-positive). c, d Comparison of probability density functions of the differences between the results obtained with AutoSpread vs. the usual SSM algorithm. This shows the small difference between both calculations. Values are displayed in log-scale of the absolute value of the difference, separated in positive (solid lines) and negative (dashed lines) values to assess any bias between positive and negative errors. e, f Comparison between results obtained with AutoSpread vs the usual SSM algorithm, but with the omission of the first regression in AutoSpread, which leads to a systematic downward bias in AutoSpread results. Same scale and line code as in c, d. Raw datasets are available at FlowRepository with IDs FR-FCM-Z2SS (MM1) [https://flowrepository.org/id/FR-FCM-Z2SS] and FR-FCM-Z2ST (HS1) [https://flowrepository.org/id/FR-FCM-Z2ST].