Computational, Systems and Developmental Neuroscience
Neural assemblies and calcium imaging data
To help us understand our data we often develop new computational models and methods. Some recent examples are as follows. We also recently reviewed computational methods for analyzing behavioral data here.
- Calcium Imaging Latent Variable Analysis (CILVA). In two papers here and here we developed a statistical model of the interaction between spontaneous and evoked activity in calcium imaging data, based on the idea that spontaneous activity can often be accurately represented by a small number of latent (unobserved) factors. Using statistical inference procedures we simultaneously derived from the data both the structure of these latent factors and neural receptive fields. This allows us to decouple spontaneous from evoked activity, providing the opportunity to study each in relative isolation.
- Detecting neural assemblies in calcium imaging data. Activity in populations of neurons often takes the form of assemblies, where specific groups of neurons tend to activate at the same time. However, in calcium imaging data, reliably identifying these assemblies is a challenging problem. In this paper in BMC Biology we compared several recently proposed assembly-detection algorithms, and demonstrated leading performance for a graph-clustering algorithm that we adapted for this problem.
- Emergence of spontaneous assembly activity in developing neural networks without afferent input. In this paper we showed that a recurrent network of binary threshold neurons with initially random weights can form neural assemblies based on a simple Hebbian learning rule, without any afferent input. This helps us understand our experimental results in this paper where such assemblies arose even in dark-reared fish. Surprisingly the set of neurons making up each assembly continually evolves in the model, a kind of representational drift.