A spatially-localized penalized matrix decomposition (PMD) designed to separate (low-dimensional) signal from (temporally-uncorrelated) noise for a wide variety of functional imaging data types (including one-photon, two-photon, three-photon, widefield, somatic, axonal, dendritic, calcium, and voltage imaging). The decomposition is applied in parallel to local spatial patches of data, making the decomposition highly scalable for large datasets. Additionally, all internal tuning parameters are estimated directly from the data -- allowing for straightforward automation of pipelines including this method. The spatially-sparse, low-rank form of the decomposition -- which is significantly compressed -- facilitates the application of downstream analyses with implementations that operate directly on the factored data (e.g. LocaNMF).
The NeuroCAAS implementation of PMD uses version 1.2 of trefide. Look for live logging in pmd_out.txt in addition to standard DATASET_NAME files.
Args:
-Motion-corrected and detrended imaging data : (npy file) data matrix formatted in shape [fov_height, fov_width, num_frames].
-Config: (yaml) a yaml file containing the following parameters:
1) fov_height: height of field of view (in pixels).
2) fov_width: width of field of view (in pixels).
3) num_frames: length of movie (in frames).
4) block_height: height of blocks to use for localized processing.
5) block_width: width of blocks to use for localized processing.
Outputs:
-Decomposition (npz file) with fields:
1) U: spatial components of decomposition [fov_height, fov_width, rank].
2) V: temporal components of decomposition [rank, num_frames].
3) scale: pixelwise normalization applied prior to decompsiiton.
4) baseline: pixelwise mean removed prior to decompsiiton.
You must login to use an analysis.