Real-time detection of motifs from semi-synthetic calcium imaging data with various noise types

The hippocampus is an important region responsible for memory formation. Some research evidence suggests that neurons in the hippocampus discharge in a sequential order to form specific activity patterns. Multiple replayed spike sequences may represent the recurrence of an individual's memory, and scientists refer to these replayed spatiotemporal sequences as motifs. Identification and online detection of these replayed spike trains are important for studying the mechanisms of memory formation. This work aims to evaluate the detection accuracy of motifs in Calcium imaging for semisynthetic data where manually identified motifs (originating from real data) are embedded into synthetic data containing various types of noise. It allows us to accurately pinpoint the timing of motif occurrences, thereby facilitating the evaluation of our method's real-time detection performance.The action potential of a neuron is typically completed within three milliseconds, while the replay time of a complete motif usually spans several hundred milliseconds. Ideally, motifs should be detected before they end in order to investigate their functions causally through intervention. For detecting motifs, we used a non-negative matrix factorization algorithm seqNMF. We conducted tests on our method's real-time detection performance for replay events. The method exhibited the ability to perform real-time detection of replayed spike sequences within a few tens of milliseconds. It was more effective in detecting large motifs than small motifs. However, we also found that the method is susceptible to increased jitter, suggesting the importance of time-precise recording for retrieving motifs robustly.