v0.2.0
This release brings a variety bug fixes, changes in default parameters, and better utilization of the UMAP algorithm.
CHANGES:
- Rosella now handles situations with 2 or less samples differently. Opting to perform a single UMAP projection
- When 3 or more sample are present then an intersection of three UMAP projections is used. These projections include a metabat distance projection on the contig coverage and variance, TNF correlation projection, and CLR coverage correlation projection. Discussions regarding combining and intersecting UMAP projections can be found here: lmcinnes/umap#206
- MetaBat 1 and 2 provide an awesome method for calculating the distance between normal distributions which has been included into rosella as distance metric. This method seems to work much better than using kullbacl-leibler divergence
- Only canonical tetranucleotide frequencies are used now.
- Should be faster in most cases due to less reliance on custom metrics for TNF UMAP and correlation UMAP