Unsupervised Neuronal Matching with Spontaneous Neuronal Activity
Abstract
To obtain deeper understandings of the brain, aligning similarly functioning neurons, or matching neurons, in different neural systems is becoming an important problem in neuroscience. A major approach for neuronal matching is stimulus- based approach, where matching is performed through similarity of neuronal activity when exerted the same stimulation. This approach, however, is experimentally time-consuming and laborious, and we are in want for a more widely applicable matching approach that possibly uses more accessible data, such as spontaneous neural activity. Here we propose a neuronal matching framework that uses the spontaneous activity. The proposed method is based on an extension of Gromov-Wasserstein optimal transport (GWOT) (Memoli, 2011), which we named Gromov-Wasserstein optimal transport with multiple distance matrices (GWOT-MD). As a test of efficacy of the proposed approach, we applied the proposed framework to calcium imaging time series of spontaneous neuronal activities of \textit{Caenorhabditis elegans} (\textit{C.elegans}). Ratios of matching with pre-identified labels between individual pairs turned out much better than chance level matching ratios. We also performed neuron label identification using the matching results and revealed that the top 5 identification accuracy turned out as good as an identification method using neuronal locations (Sprague et al, 2024).