anchor cells should be internally consistent

Nanostring-Biostats / InSituType

An R package for performing cell typing in SMI and other single cell data

Other

29 stars 11 forks source link

Closed patrickjdanaher closed 2 years ago

patrickjdanaher commented 2 years ago

Implement a method to throw out anchor cells that are outliers from the rest.

Ideas:

call a centroid, get candidate anchors' distances from centroid, remove those with high outlier distances. (scaled loglik and cosine both decent candidates)
hclust on their correlation matrix. Cut at varying heights, seeking a cut that 1. seems to be a majority and 2. has high cos/llr scores.

patrickjdanaher commented 2 years ago

Plan:

calc new centroid of anchors
calc cosine distance from new centroid
throw out cells whose new cosine distance is lower than the preset threshold (or throw out lowest 10%)

patrickjdanaher commented 2 years ago

Other anchor bugs:

[x] in "anchorslist" creation, should omit cells with NAs for the column in question
[x] insufficint anchors arcgument must be exposed to insitutype
[x] must allow for no anchors to survive after applying insufficient_anchors argument. Maybe write anchors=NULL when this happens?
[x] Need to get llr from 2-3rd best cosines from each row.

patrickjdanaher commented 2 years ago

Implemented in a fairly weak form.