News#
MIRA version 2.0 - 06/20/2023#
With our next publication in Nature Communications, we introduced a new batch effect correction method which is purpose built for multiomic analysis. The algorithm, called CODAL, uses a novel objective function to “disentangle” technical effects and maximally preserve biological variance. This makes the CODAL algorithm especially well suited to studying the effects of perturbations on single cells, where the biological change induced by the perturbation could be obscured by technical noise. In benchmarking tests, the CODAL algorithm produced results on-par with current state-of-the-art supervised methods, while requiring no prior knowledge about the sample.
With the CODAL update, we completely re-vamped the hyperparameter tuning method to be 10-20x faster, more robust, and highly parallelizable. Of note, we introduced the gradient_tune method, which uses gradient descent optimization to automatically select the hyperparameters for the model, reducing the time taken for hyperparameter selection many-fold.
Coming next, the Regulatory Potential modeling section of the package is due for a major update which will dramatically increase its speedy, scalability, and interpretability.
MIRA version 1.0 - 03/15/2022#
This documentation website will be launched alongside the first official version of MIRA! The models/statistical code is stable, and most new features will be devoted towards accessing information more easily, decreasing runtime, and adding more flexibility to plotting functions. Releasing version 1.0.0 reflects that this code embdoies what we set out to design, and represents a comprehensive base of features on which users can build multiomics analyses. The biggest changes with this release are:
Multiprocessing and tensorboard support for topic model training
New topic model trial pruner which saves time and improves results
Multiprocessing for RP model training
Facilities for visualizing regulatory potential models
Bug fixes and tweaks to streamplot functions
Added seeding to topic model tuning and terminal point selection to facilitate deterministic analyses
Accessors for various attributes from AnnDatas (motif hits, TSS-peak distances, etc.)
Extended support for MIRA streamgraphs and lineage inference to velocity experiments via CellRank
Added pre-built TSS and chromosome size annotations for mouse (mm10) and human (hg38)
The only breaking changes are some plotting functions had an add_legend parameter, while others had show_legend. We changed this so that all plots have show_legend, consistent with Seaborn conventions.
Code Ocean Capsule - 03/15/2022#
We packaged our analysis of the SHARE-seq skin dataset (Ma et al. 2020) as a code ocean capsule, which enables users to check out the code used to generate figures and conduct the analysis.
New website! - 03/15/2022#
Welcome to MIRA’s new documentation website! We hope you find it informative and aesthetic. Future additions to documentation and expansions to the MIRA analysis suite will live here.