Gseapy

Fixed a critical bug in NES null distribution calculation in the Rust code. 162

And FDR values is affected in when NES < 0 for gseapy version (0.12.0, 0.12.1).

Please update at least to v0.13.0 to get the correct FDR values.


minor

- improve output files (sort by abs(NES) in descending order)
- generate output gene signature file (gmt) used in the calculation


**Full Changelog**: https://github.com/zqfang/GSEApy/compare/v0.12.0...v0.13.0

This is the first Rust binding of GSEApy

NOTE
This version have bugs in NES calculation
**Please update to at lease v0.13.0 to get the correct FDR values!!!**


Summary
- This version is re-written in Rust, which solve major problems such a memory cost and speed.
- Rust Binding of GSEApy is at lease 3-fold faster than Numpy implementation and 4 times less memory cost (Prerank module). 134, 142
- New output format are used in this version and will be adpoted in the future releases.
- All API remain the same to the older version of GSEApy(< v0.10.8).

New features:

- Rust binding of GSEApy is much faster than the Numpy version (<=v0.10.8)
- The `gene_sets` argument in `gsea`, `ssgsea`, `prerank`, now support multiple library inputs, e.g:
- list: ['KEGG_2016,'KEGG_2021_Human']
- dict: {'term_1': ['gene1', 'gene2,' ...], ...}
- multiple libraries seperated by comma: "KEGG_2016,KEGG_2021_Human"
- single libraries: KEGG_2016
- or gmt file: "KEGG_2016.gmt"
- Note: it will save you a lot of time if you put multi-values at once, instead of run each seprately.

- dotplot, heatmap, ringplot improvement quite a lot !


Deprecated

- argument `processes` are now renamed to `threads`

- Fix t_test statistics, 141
- Fix a critical bug in phenotype permutation percedure (the permutation for null distribution is accidently not correct for phenotype shuffling proceducre) 140. Please update your results using v0.10.7 (only affects `phenotype permutation`)
- Add odds ratio output to Enrichr output
- minor docs improvement

Fix Enrichr URL error. 110 , 111

This is a major update.

* Fixed a critical bug in NES calculation 105.
* A few minor improvements for a better and cleaner interactive computing experience
* add a new ranking method: genes with absolute value of signal to noise ( `abs_signal_to_noise` or `abs_s2n`)
* change cache directory from `~/.gseapy` to `~/.cache/gseapy`.

* Drop dependency of bs4
* Performance improvement for large dataset: use joblib instead of multiprocessing 92, 70
* Fixed random seeding. Now FDR results are reproducible 95, 70
* Improve heatmap ticks layout