kevinwang09 / cpop Goto Github PK

Perform sort before pairwise_col_diff

plot_top_coef with type = "bar" is stacked rather than compared.

CPOP uses glmnet as the default backend. However, in the case that only one feature can be selected into the model, this causes issues for the cpop2 function to further perform calculations. The reason for this is due to glmnet
https://stackoverflow.com/questions/29231123/why-cant-pass-only-1-coulmn-to-glmnet-when-it-is-possible-in-glm-function-in-r

A solution to this would be using that single selected feature in a more classical model fitting procedure, e.g. glm(). This ensures that a result is given to the user. However, a warning message is necessary here because this edge case detaches itself from the original CPOP manuscript and the intended use.

The k features with smallest normalised distance between en1 and en2 are selected into the model.

For top::top, add an extra functionality that adjusts the alpha parameter in iterations to select more features. Alternatively, if the desired number of features has been exceeded, then we will break the for-loop.

Since the TOP method requires two datasets for training, we propose that the predictions from both models can be averaged to produce a more sensible prediction for a third data.

Also, since we are talking about the model averaging, can we average out the model predictions using different feature sets (top1_iterate function output).

Currently, all the pairwise features will be created before predictions can be made. This can be simplified by configuring the final glmnet model so that only relevant features are kept, and thus reducing the need to parse through large data object when making external validation predictions.

Network plot:
Node size: t-stat of gene
Node color: sign of the t-stat of gene
Edge color: sign of the ridge regression coef
Edge size:: magnitude of the ridge regression coef

Currently, there are a lot of inputs to each function. This is messy to code. One option is to use some slots to manage this.

Some references:
http://adv-r.had.co.nz/S3.html
https://stuartlee.org/post/content/post/2019-07-09-s4-a-short-guide-for-perplexed/
https://bioconductor.org/packages/release/bioc/vignettes/MultiAssayExperiment/inst/doc/MultiAssayExperiment.html