Besides loading gene expression already normalised, it would be great to normalise by our own means.

Check function fnRowCallback for implementation (possibly called rowCallback in Shiny's DataTable). Also, check this example.

This may depend on the analyses for different dataset types. For instance, an analysis that works with gene expression or junction quantification. However, there are times where two different datasets of the same type may be loaded (for example, two datasets of junction quantification obtained through different methods).

One such place that could use text suggestions is when asking the user for folders where data is stored (don't forget that the filesystem separator differs by OS and can be retrieved in R using the variable .Platform$file.sep).

• Selecting folder
• Subset expression (suggest dataset columns)

• Exon skipping
• Mutually exclusive exons
• Intron retention (not available using only junction read counts)
• Alternative 3' splice site
• Alternative 5' splice site
• Alternative first exon
• Alternative last exon
• Tandem UTR

The user might want to see all events.

Allow to plot the time series in days, months, years or custom time ranges.

Instead of having a tab to edit groups, it would be best to allow group editing only when asking a user for groups. This way, it would be much more intuitive to understand what the groups are for with no more explanation (in the current method, people need to understand what the groups are for before doing analysis).

The progress bar value doesn't reflect the loading state of the files. When the files are completely loaded, it reaches around one third of the progress bar.

Wilcoxon test warns that it cannot compute exact p-values with zeroes or ties in the console. These warnings should be in the user interface instead.

• Show warnings in the interface
• Describe the error

See also #7 and check biomaRt and Ensembl API.

This may include drawing the supporting reads.

• mRNAs with splicing features
• Proteins should contain domain information
• Ask LG for more info on protein domain comparison

To help with this, see the JavaScript library autosize.

This could be a user setting.

The groups should be easily (un)selectable by using a master checkbox (which is actually there right now but that does nothing). This could be solvable by calling the JavaScript function after the checkbox is generated (using a DataTables' callback).

Easy to do it with the DT R package, see #151.

• Heatmap
• Allow to plot heatmap with the points of other analyses

Right now, the fuzzy index is given on a word basis based on alphabetical sort with an inappropriate cutoff.

Compare the resulting genes with pathways of interest.

• Create groups by selecting a data column and using a GREP expression on that column.
• Add a warning on failure

There are some custom functions that have been included in the development version of Highcharter. As soon as this version is released in CRAN, there's no need for them to be in this project.

Instead, it could be better to hide these tests or simply state why they can't be performed.

GSEA allows to study over-represented classes of genes in a set of genes.

Subramanian et al. 2005. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. PNAS, 102 (43) 15545-15550

This could be achieved by selecting a colour for each created colour in the groups section. The colours used in there would then be used in all plots when representing that same group.

This may reveal a bit problematic if there are a great number of groups.

The number of cores to use could default to 1 (for safety reasons) or maybe use 2 or 4 depending on cores available… is it possible to check available cores in R?

This should be a setting the user could modify (see #80).

This would be easy with Highcharter using the Annotations plugin since the jbkunst/highcharter#103 bug fix. So… should we wait for this to be implemented or go with a workaround?

It can be a bit slow to calculate all statistical tests for the events. For this reason, this action should probably be initiated by the user.

The rank/sort feature for the user is easy if this is done in a Data Table.

Currently, this box is too big (if the elements within it are big) and may be placed outside the browser window which doesn't allow the user to read the suggestions.

If not possible, an alternative to resizing the box is to at least force the left side of the box inside the browser window.

According to NM, we could create a lot of random points from two different beta distributions and compare them using the Mann-Whitney U test.

Depending on the combination of the groups used, the plot can be empty. Also, hiding a certain group (usually the first one) can hide all other groups as well.

This is needed to properly document, export and test functions. Functions inside subdirectories in the R directory are not considered by devtools.

This will allow faster loading of files next time the program opens. By using the exact same name (excluding extension), it'd be easy to compare which files have RDS equivalents or not.

Right now, it only tries to get what's inside the path without giving any error.

• Check if path is not to a file (else warn the user) (2f71fa1)
• Check if folder exists (else warn the user) (2f71fa1)
• Only suggest folders in suggestions instead of showing both folders and files (43e9a90)

• Separation with k-means or a line
• Validation with Fisher's exact test

To check if a data frame is NULL, use this (source):
condition = "(typeof input.df !== 'undefined' && input.df.length > 0)"

These clusters could then be used in survival curves and other analyses.

• Separate based on user-created line (maybe validated with Fisher's exact test)
• Separate by k-means / PAM / CLARA
• Assist methods by showing optimal number of clusters (using total within sum of squares, silhouette width and gap statistics)
• Validate with cluster validity methods
• Plot clusters in PCA using the polygon series
• Create groups based on clusters

See also cluster analysis in R (1 and 2) and cluster validation using cIVALID