Skip to main content
. 2020 Jun 5;9:e55913. doi: 10.7554/eLife.55913

Figure 6. pea3 smFISH as an Example of Data Integration Across Imaging Modalities.

(A) Maximum z-projection of a two-color stack of pea3 smFISH (yellow) and the lyn-EGFP membrane marker (magenta). Scale bar: 10 µm. (B) Results of SVR regression on pea3 spot counts using TFOR and CFOR shape features as well as cell centroid coordinates of registered primordia as input. Each blue dot is a cell, the diagonal gray line reflects perfect prediction and blue arrows at the border point to outliers with very high spot counts. On training data, the regressor's explained variance ratio is 0.462 ± 0.011, on previously unseen test data it achieves 0.382 ± 0.019. (C–D) Consensus tissue maps of pea3 expression generated directly from the pea3 smFISH dataset (C) or from the full atlas dataset based on SVR predictions of spot counts (D). Note that the prediction for the entire atlas preserves the most prominent pattern – the front-rear gradient across the tissue – but does not capture the noisy heterogeneity among follower cells observed in direct measurements.

Figure 6.

Figure 6—figure supplement 1. Spot Detection and Cell Shape Embedding for pea3 smFISH Data.

Figure 6—figure supplement 1.

(A–C) Maximum z-projections of a two-color sample showing the lyn-EGFP membrane marker (A), the pea3 smFISH probe (B), and the results of automated spot detection with red rings denoting detection events (C). Scale bars: 10 µm. (D) Zoomed view of the region in the yellow box in (C). Scale bar: 2 µm. (E) pea3 smFISH spot counts for each cell, both from measured data (blue) and from predictions across the entire atlas dataset (purple). The left shows averages across primordia, which closely match those reported previously based on a different spot counting method (Durdu et al., 2014). The individual cell counts on the right show that there is a long tail of cells with extremely high counts, which as one would expect is not captured in the SVR predictions. (F–H) Comparisons of three important cell shape variables between fixed smFISH samples and live samples (here the set of live samples containing only the membrane label; N = 24, n = 2310), showing no significant difference.