There are a few different things that have been changed here:
significant speed boost in sleuth_fit by doing the measurement error and beta covariance calculations with the full matrix rather than by row
significant memory improvement by removing any environment associated with formulas supplied by users.
added functionality for users to supply their own shrinkage functions in sleuth_fit using the shrink_fun argument. The default sleuth shrinkage function is a new function called basic_shrink_fun. I also implemented limma's shrinkage function in a new function called squeezeVar. This should address concerns raised by #173.
changed the API for log_transform and added one for TPMs (tpm_transform), so that they can take a single size factor when normalizing bootstraps, or can use a vector of size factors equal in length to the number of samples when transforming the full matrix. This allows for sleuth-ALR to separate the normalization and transformation steps.
I have also fixed various small bug fixes. I'll update this description when I've merged in more:
I fixed #190 by bringing back gene_from_gene. This incorporates the same patch included with pull request #192.
Replaced outdated eval statements in the shiny scatter plot panel with rlang::eval_tidy to fix #194
Improved sanity checking in sleuth_results to better handle when gene_mode and pval_aggregate are both TRUE, to fix #202
To-dos:
[x] add download button for LRT table in shiny to fix #62 (and check for other missing buttons)
[x] add a better error message to alert users to the edge cases brought up in #120 and #175
[x] fix dependency error seen in #135
[x] fix bug when users supply factors when loading kallisto results, seen in #197
[x] fix bug when subsetting kallisto, seen in #204
[x] fix heat map issues in #207 and #208
[x] fix gene view error seen in #209 (I think this has been fixed by 5b4a3308d69254c0fab838b188cd0a6bb75da54c)
[x] fix typo in sleuth_prep documentation mentioned in #212
Hi @pimentel,
There are a few different things that have been changed here:
sleuth_fitby doing the measurement error and beta covariance calculations with the full matrix rather than by rowsleuth_fitusing theshrink_funargument. The default sleuth shrinkage function is a new function calledbasic_shrink_fun. I also implemented limma's shrinkage function in a new function calledsqueezeVar. This should address concerns raised by #173.log_transformand added one for TPMs (tpm_transform), so that they can take a single size factor when normalizing bootstraps, or can use a vector of size factors equal in length to the number of samples when transforming the full matrix. This allows for sleuth-ALR to separate the normalization and transformation steps.I have also fixed various small bug fixes. I'll update this description when I've merged in more:
gene_from_gene. This incorporates the same patch included with pull request #192.evalstatements in the shiny scatter plot panel withrlang::eval_tidyto fix #194sleuth_resultsto better handle whengene_modeandpval_aggregateare bothTRUE, to fix #202To-dos:
sleuth_prepdocumentation mentioned in #212