Fig. 5: Ablation results.

a, We tested six different ablations across all simulation conditions: no drift correction, rigid motion correction only, no template deconvolution, no reclustering after template deconvolution, no cross-correlogram (CCG) information in deciding splits and merges and no feature deconvolution for the final clustering step. b, Simulation results quantified as changes in the number of misses (top) and changes in the number of false positives (bottom).