Neural data science: from raw neuroscience recordings to scientific discoveries (Pengcheng Zhou - Columbia University)

With the adoption of increasingly powerful neurotechnology, novel quantitative approaches for understanding these new data sets of massive size and complexity are in increasing demand. In this talk, I will present two novel methods for extracting individual neuronal signals from in vivo calcium imaging data collected in two unique experimental setups. The first technique uses microendoscopic lenses to image previously inaccessible neuronal populations deep within the brains of freely moving animals. Unfortunately, it is computationally challenging to extract single-neuronal activity from its data due to the large contaminating background fluctuations and high spatial overlaps intrinsic to this recording modality. The second recording combines the conventional 2-photon calcium imaging and electron microscopy (EM), providing arguably the most powerful current approach for connecting function to structure in neural circuits. However, no automated analysis tools yet exist that can match each signal extracted from the CI data to a cell segment extracted from EM; previous efforts have been largely manual and focused on analyzing calcium activity in cell bodies, neglecting potentially rich functional information from axons and dendrites. For both of these two valuable experimental recordings, we model the observed imaging data using a constrained nonnegative matrix factorization (CNMF) framework and propose efficient iterative procedures for solving the resulting optimization problems. The methods were implemented into open-sourced packages (CNMF-E & EASE) and CNMF-E has been widely adopted by the neuroscience community.