Last week, Zac Spalding (2nd year BME PhD student) presented Gallego-Carracedo et al.’s 2022 paper investigating the relationship between latent dynamics of neural populations and local field potentials (LFPs) during movement. [cont.]
Thank you to the authors at Imperial College Bioengineering for your work! cc: Cecilia Gallego-Carracedo, Matt Perich, Raeed Chowdhury, Lee E. Miller, Juan Álvaro Gallego
❔3️⃣: Other studies find that latent dynamics are well-modeled by nonlinear dimensionality reduction methods (variational autoencoders as opposed to PCA, for example). Would you expect different results in this study when using nonlinear dimensionality reduction?
❔2️⃣: As latent dynamics – LFP correlations profiles are expected due to biophysical network architectures, do you expect them to be preserved across task types, assuming that the recorded area is active during this task?
❔1️⃣: How was the decoding performance of decoders trained on latent LFPs (as shown in figure 3, supplement 6)? Did they show comparable accuracy to decoders trained on full LFP signals?
🤍3️⃣: The latent LFP analysis (figure 3, supplement 6) is interesting, as it shows that low-dimensional representations of the LFP are still informative to population level activity.
🤍2️⃣: Figure 1 provides a great visual introduction to the paper by clearly showing how biophysical connections between neurons give rise to both LFPs and latent dynamics across a population.
🤍1️⃣: Correlation analyses are well-formulated with controls or references from tensor maximum entropy and inter-trial period data.
They find that latent dynamics and LFPs are correlated in a frequency-dependent and region-specific manner. This 🧵explores our thoughts (🤍& ❔) elifesciences.org/articles/73155
There is a frequency-dependent association between the local field potential and the coordinated activity of populations of single neurons, which remains constant during different aspects of behaviour...