- i.e. if there is/was a singular conclusive message of wide interest that could be pointed to, then perhaps that could point to an alternative title direction
- but this title is still likely okay, especially with the results and figure data so far being the same as the biorxiv papers
- domain maps. the non-random structure/overlap, area/lobe specific characteristics
- our work shows substantial overlap in the domain mosaic between recordings (at 1x mag with optimized information extraction) suggesting underlying, cortical organization-- structure. structured functional motifs in between the microscale (cellular) and macroscale (tissue level). What is left in between? Cell(s) with connections. Assemblies of cells. Cell groups. Functional units/motifs. Modules. Domains. Functions. Objects. Things.
- domains or components per recording, spatial characteristics of the domains, diam
- jaccard overlap of the domain or region borders between sequential recordings or different animals
motif (wn, noun)
: a design or figure that consists of recurring shapes or colors, as in architecture or decoration
* Q: Would the isofl data help with this manuscript? Or maybe the idea of integrating any new material is too much now
* Q: Should an information rate per recording be reported?
- the data amounts in and out of optimized vs non-optimized imaging sessions
- the avg signal power of data in and out of optimized recordings and pipeline
* Q: figure 2D, What is the value of the mean neural component curve when it levels off? Maybe around 1000 s or 16.6 min? seventeen minutes.
<!-- * Q: How many bits of the cameras ADC are needed to encode our calcium signal variance? What about for calcium signal variance+hemodynamic variance+vascular artifact variance all together on the same monochromatic channel? When we examined this once before, I think we figured out with the range of useful variance in our recordings for desired and undesired source signals (dim hemodynamics and tissue autofluorescence vs neural signals vs movement artifacts) that it is definitely more than 8bits but less than 14bits. -->
* the general solution to multichannel source signal separation, independent component analysis, can optimally recover neural signal content in recordings of neuronal cortical calcium dynamics
- when captured at a magnification factor of one and at a sampling rate of 1.5×10⁶ pixels per one-hundred millisecond frame for seventeen minutes
* a set of spatial and temporal metrics can be used to build a random forest classifier which separates neural activity and artifact components automatically at human performance
* a functional segmentation of the mouse cerebral sheet, providing a map of 115 domains per {x} cm²-mouse-neocortical-hemisphere from which extracted time courses maximally represent the underlying signal in each recording
* Q: Should statistics of our time series be commented on more?
- i.e data density/sampling over relevant scales be emphasized?
- Single plane sensor array pointed at a single living subject
- widefield, pixelsize, reproduction ratio
- [ ] def 'limited references' better
- the statistical model baseline
- statistical power of multivariate *sufficiently dense* sampling within a space-time window (scale | frame of reference | local viewport | field of view)
- the dense sampling in space and time **from single individuals** is key. The gaussian baseline estimate from analyzing movies of sufficient duration. Compared to the group averaged studies and less-than-optimal baseline assumptions that are typically utilized in most studies and applications using either unsupervised or supervised ML implementations
* [ ] Clarify 'compare control data recorded in glial cell reporter and non-fluorescent mouse lines...'
* [ ] Possibly replicative information about 'segments independent functional units' and 'produce segmentations of the cortical surface'? (unless 'surface segmentations' are the cortical areas containing the unitsin which case we have two different uses of 'segmentation'?)
* [ ] 1. First usage of term 'data-driven method' is not till near end of fourth paragraph, but should be clearly made associated with any introductions of ICA earlier or unsupervised ML learning techniques in general so that better contrast is made with the supervised ML classifier methods, as we should carefully do in the abstr as well rw
* [ ] 2. because of lack of spatial density sampling
- def and expand this more
- can add same is true for many or most other applications of ICA or other eigendecomp routines in neuro- (and proabably most fields?)
- e.g. [^Mukamel:2009] ICA used with 2P laser scanning calcium imaging time series at microscale (cellular) level. Much lower data ingests
* [ ] 3. IC model requires Gaussian baseline and independent message source
- must have one gaussian component
- many other investigations do inter-subject grouping
- message source independence
* [ ] 6. def What is baseline
- the controls are non/less-time variant tissue fluoresence vs high dynamic range neuronal calcium signals
* [ ] 8. Q: What is the median *significant* pixel count for a vascular artifact components vs a neural components?
* [ ] 9. Distributions of component shape, vacular vs hyperellipsoid neural blobs, small domains, possible types of cellular localization of calcium src
* [ ] 10. Then leads to: Where the calcium flux comes from, multi spots, small comp in opp hemisphere from the larger singular src blobs, axon traj or max prob of tissue src origination
* [ ] rw 'components that are unsorted and often flipped'
* [ ] rw 'mean time series is pre-subtracted from the array before SVD'SVD used before defined in next paragraph. The mean signal effect is removed with the pre-whitening/sphereing step of SVD
* [ ] rw, clarify Map creation domain size as in: 'This map was then further processed to get rid of domains smaller than 1/10th the mean domain. Any domain smaller than this size'
* [ ] rw, clarify 'Olfactory bulbs were included in map generation, but domains were highly variable, and were excluded'
* [ ] check sentence 'same number of domains or regions as the original map, n, and shares the same cortical mask as the original map. To create this map, n points were distributed randomly across the cortical mask. To turn these points into regions, the voronoi diagram was created using the scipy spatial package35'
* [ ] check sentence 'When comparing maps generated from different animals, the optimal alignment was calculated by shifting the second map up to 100 pixels in any direction. The optimal direction was determined by maximizing the Jaccard overlap.'
