Releases: az7jh2/SDePER
SDePER v1.6.2
SDePER v1.6.1
Bug Fixes:
- Changed the dependency to the headless version of OpenCV to avoid installation issues and since no GUI functionality is required.
SDePER v1.6.0
Updates:
- The
imputation
module has been completely rewritten, significantly improving contour finding and grid generation for enhanced performance and accuracy (#1).
SDePER v1.5.0
Updates:
-
The optimization of cell type proportion
$\theta$ is skipped if the initial value of$\theta$ indicates the presence of only one cell type in the spot. -
When predicting cell type proportions utilizing the CVAE latent space, the values of the CVAE latent space are now directly used instead of PCA embeddings for proportion transferring.
-
The default number of hidden layers has been changed from 2 to 1.
SDePER v1.4.0
Updates:
-
Added single cell augmentation feature for scRNA-seq reference data.
-
Implemented failure steps for handling irregular
$\theta$ values after optimization. -
Added support for transferring cell type proportions based on PCA or UMAP embeddings of latent space, or directly on the original latent space in cell type proportion prediction using CVAE.
-
Introduced caching in GLRM to accelerate computations by storing calculated likelihood values, reducing duplicate calculations. Note: This feature is disabled by default due to potential optimization failures caused by unknown reasons (#5).
Bug Fixes:
-
Resolved issue with spot name inconsistencies when spots are filtered out if cell type proportions predicted by CVAE were used for
$\theta$ initialization in GLRM modeling. -
Fixed bug causing errors when plotting CVAE loss during training in the absence of validation data.
SDePER v1.3.1
Updates:
-
Added a step to remove mitochondrial genes during preprocessing.
-
Introduced a PCA plot for visualizing the latent space and added density estimation based on PCA in diagnostic figures.
SDePER v1.3.0
Updates:
-
Introduced prediction of cell type proportions utilizing the CVAE latent space. Currently, the proportions are transferred from the scRNA-seq condition to the spatial condition in latent space. Then the predicted cell type proportions are used as initial value of
$\theta$ for GLRM modeling (#13). -
Reused
$\theta$ and$e^{\alpha}$ estimations from stage 1 of GLRM modeling for initializing stage 2 (#12). -
Increased the weight of spatial spots and scRNA-seq cells in CVAE training against generated pseudo-spots.
-
Added support for retaining only highly variable genes in the spatial data. By default all genes are retained.
-
SDePER options are written to a text file within the diagnosis folder, and only DE genes are retained in the CVAE-transformed data during saving if command-line option
--redo_de
istrue
. -
Decreased the default number of command-line option
--n_pseudo_spot
to100,000
.
Bug Fixes:
- Resolved a bug where errors occurred during diagnostic UMAP drawing if only cell type markers were provided, and no scRNA-seq cells were available.
SDePER v1.2.1
Updates:
-
Implemented a GitHub Action triggered by new release publications to test the package installation (#10).
-
Added a diagnostic UMAP plot of raw data before platform effect removal using CVAE. Also included new diagnostic plots depicting CVAE training loss.
-
Changed the default value of the
--n_marker_per_cmp
command-line option to 20. -
Added three command-line options:
--use_batch_norm
,--use_spatial_pseudo
and--cvae_train_epoch
.
SDePER v1.2.0
Updates:
-
Revised the
cvae
module, implementing several updates including (#4):-
Integration of Batch Normalization into the CVAE training process.
-
Inclusion of a logarithmic transformation in the preprocessing of gene expression data for CVAE input.
-
Generation of “pseudo-spots” under spatial conditions through the random combination of spatial spots.
-
Addition of two command-line options:
--n_pseudo_spot
and--num_hidden_layer
. Also adjusted the default value of--cvae_init_lr
.
-
-
Relocated all code related to generating diagnostic figures to a new module,
diagnosis_plots
. Additionally organized the output figures into a folder named diagnosis within the output path (#6).