Currently the lineage registration plugin registers two embryos and computes the angle between two paired cell divisions.
But maybe the directions of cell division are not as much conserved with respect to a global coordinate system but are better defined if compared in a local coordinate system that uses the cell division directions of the parent cells for reference.
Steps
[x] Find a way to compare cell division directions with respect to the parent cell divisions direction
[x] See if the resulting angles are better preserved as with the current approach
Results are very inconclusive, probably not better
[ ] ~Find a way to integrate this into the lineage registration plugin~
Idea
Let a1 be a cell division direction in embryo A. Let a2 denote the parent cell division direction, and a3 denote the grant parent cell division direction in embryo A.
Let b1, b2, b3 denote paired cell division directions in embryo B.
The points a2, -a2, a3, -a3 could serve as landmarks and be registered against b2, -b2, b3, -b3, the resulting transformation could be used to transform a1 and b1 into a common coordinate system, where one could measure the angle between the to cell division directions.
Results
Two methods of local registration for cell division direction comparison have been developed
As described in Idea.
Generate Landmarks from the siblings of the cell that divides. Compute the average position of the sister cell during cell division timepoint, average position of the "cousin" cells, average position of the "second cousin" cells. This gives as 3 landmarks, enough to be used to compute a spatial registration.
The average angle between cell division has been computed an compared. The average angle has been computed in bins over time and plotted. For some embryos there seem to be improvements:
But these improvements very much depend on the meta parameters, they are not consistent between macrostomum and platynareis.
A quality measure was implement, which was motivated by tree edit costs. This quality measure indicated different optimal meta parameter.
When consistency or lineage registration was tested. The local registration approach didn't yield any improvement.
Conclusion: The potential of local spatial registration is still inconclusive. It seem to improve some measures under certain conditions but not at all consistently and it doesn't improve the end result. It might help registration of later timepoints, but we probably fail earlier in lineage registration. We would need a more robust approach to leverage the potential of the local spatial registration.
There might be a pattern in division angles between consecutive cell division, that is picked up by the approach and leads to the reduce local angle on later timepoints.
Multi Embryo Lineage Registration Consistency Check:
A useful method for checking the consistency of lineage registration across multiple embryos was developed:
Currently the lineage registration plugin registers two embryos and computes the angle between two paired cell divisions. But maybe the directions of cell division are not as much conserved with respect to a global coordinate system but are better defined if compared in a local coordinate system that uses the cell division directions of the parent cells for reference.
Steps
Idea
Let a1 be a cell division direction in embryo A. Let a2 denote the parent cell division direction, and a3 denote the grant parent cell division direction in embryo A.
Let b1, b2, b3 denote paired cell division directions in embryo B.
The points a2, -a2, a3, -a3 could serve as landmarks and be registered against b2, -b2, b3, -b3, the resulting transformation could be used to transform a1 and b1 into a common coordinate system, where one could measure the angle between the to cell division directions.
Results
Conclusion: The potential of local spatial registration is still inconclusive. It seem to improve some measures under certain conditions but not at all consistently and it doesn't improve the end result. It might help registration of later timepoints, but we probably fail earlier in lineage registration. We would need a more robust approach to leverage the potential of the local spatial registration.
There might be a pattern in division angles between consecutive cell division, that is picked up by the approach and leads to the reduce local angle on later timepoints.
Multi Embryo Lineage Registration Consistency Check:
A useful method for checking the consistency of lineage registration across multiple embryos was developed:
https://github.com/mastodon-sc/mastodon-tomancak/blob/multi-embryo-registration/src/test/java/org/mastodon/mamut/tomancak/lineage_registration/MultiEmbryoRegistration.java
It should be turned into a plugin.