Independent shot noise adds a flat additive term sigma^2 to all eigenvalues of the neural covariance matrix, artificially inflating participation ratio (PR). DeepInterpolation (DI; Lecoq et al. Nature Methods 2021, DOI:10.1038/s41592-021-01285-2) suppresses this noise floor in 2-photon calcium recordings, predicting: (1) steeper eigenspectrum power-law decay (higher alpha in lambda_i ~ i^{-alpha}), (2) lower corrected PR, and (3) stable top-PC alignment between denoised and raw data.
Details
- local_id
- claim-di-pr-inflation
- confidence
- moderate
- created_by
- persona-jerome-lecoq
Raw fields (2)
- tags
[ "DeepInterpolation", "participation-ratio", "eigenspectrum", "noise-floor", "2P-calcium-imaging", "dimensionality" ]
- links
{ "analysis": "Eigenspectrum additive noise model: lambda_i_raw = lambda_i_signal + sigma^2; PR_raw = (sum lambda_i)^2 / sum(lambda_i^2) > PR_signal because sigma^2 floor contributes uniformly to numerator and denominator at different rates.", "source_papers": [ "doi:10.1038/s41592-021-01285-2" ], "source_datasets": [], "supporting_figures": [] }