Calcium imaging is a key method to record from identified neurons in the living brain. To infer spiking activity from neuronal calcium signals, we have previously developed CASCADE, a method based on supervised deep learning1. To train the algorithm, we assembled a database of simultaneous calcium and electrophysiological recordings from the same neurons in several brain regions in mouse and zebrafish1. Here, we provide additional datasets and analyses. First, we now include ground truth datasets recorded from excitatory and inhibitory neurons of mouse spinal cord. A CASCADE model for spike inference that was trained on previously available ground truth generalized well to calcium imaging data from spinal cord. However, a model specifically trained on spinal cord data improved spike inference and recovered absolute spike rates and burst events more reliably. Second, we analyze a publicly available ground truth datasets focused on GCaMP8 2. We analyze how well models trained on previously available ground truth data generalize to GCaMP8 data. Furthermore, we investigate how fast onset kinetics of GCaMP8 results in systematic temporal shifts of inferred spikes compared to ground truth spikes. Finally, we analyze how well single spikes can be inferred from GCaMP8 data. Together, we analyze what aspects have to be considered for spike inference from calcium imaging data recorded with GCaMP8.These updates demonstrate how CASCADE can be continually extended using newly available ground truth datasets for optimized spike inference from calcium imaging experiments. References: 1 Rupprecht et al., Nature Neuroscience (2021). 2 Zhang et al., Nature (2023).