Enhancing flexibility, removing heuristics and changing knn construction

[tluquez]

Subsample based on the percent of cells

I added support for the user to specify the percent of test cells rather than the absolute number to downsample to.

Split test based on percent of cells

I added support for specifying the percent of test cells rather than the absolute number. This solves problems when donors have fewer cells than the absolute number but enough to split on percent.

For reproducibility, I also added support for specifying the seed.

Computing modularity safely

A frequent error was computing modularity for a k outside the range of the merges in the walktrap object. I added a safe way of automatically computing modularity for every k and a column with the differences in modularity to help in a posteriori k selection.

From correlation-based to Euclidean distance

For some matrices, the correlation distance had problems ensuring all the entries were positive. I changed the distance to Euclidean. Unlike 1 - correlation, It does not assume linearity nor normality and does not violate the triangle inequality.

Enhanced user control

Training many scHPF models is time-consuming. I added support so that the user can specify 1) whether to rewrite those models and 2) an apriori k. This is much more convenient.

Miscellaneous

• Export modularity table for a prior k determination.
• Plot all k values in the walktrap modularity plot.
• Shuffle code around to make it more modular.
• Added more messages to facilitate debugging and user feedback.
• Changed the functions to retrieve the spectra so that factors can be individually recognized in case the user wants to use a specific one.