Figure 6.

Validation of AutoML methods on modified 2p calcium imaging data. (A) Examples of ROIs extracted from 2p calcium imaging modified to resemble 1p data and to simulate ROIs that should be excluded from the analysis. Example trace data (top) and spatial footprints (bottom) of simulated negative ROIs that were modified using various methods: “spatial” (spatial footprints were upscaled or downscaled beyond the range of typical neuronal size in the dataset), “trace” (traces with Gaussian noise added), “both” (where both “spatial” and “trace” modifications were made) or merged” (where two ROIs are incorrectly identified as a single ROI by combining the spatial footprint and trace of two separate cells). (B) Confusion matrices of the TPOT (left) and AutoSklearn (right) test predictions on the 2p simulated ground truth dataset (where the test set size is 1,080 ROIs).