proseg-live 3.2.0.3__tar.gz

proseg_live-3.2.0.3/CHANGES.md

@@ -0,0 +1,189 @@
+
+# 3.2.0
+  - Fix a bug when regressing out z-axis tilt.
+  - Add support for reading MERSCOPE parquet files.
+  - Fix two minor (likely inconsequential bugs) in the sampler.
+  - Tweaks in the factorization scheme to better accommodate whole transcriptome data.
+  - Calibrate priors to better account for high transcript density assays.
+
+# 3.1.1
+  - Fix some inaccuracies in how posterior mean estimates are computed.
+  - Check that the output directory exists before running inference.
+  - Retain transcript quality values and output them with the transcript metadata.
+  - Fix a bug causing excessive memory usage when writing transcript metadata
+  - Set some additional metadata in the spatialdata geoparquet files
+
+# 3.1.0
+  - Fix a significant bug in the factorization scheme which degraded segmentation quality on data with large gene panels.
+  - Extensive optimization work leading to around 1.5x-4x faster run times and 5-10% less memory usage. Changes include:
+    - Redesign sparse count matrix data structure (now using csr with B+-tree rows, and row-level read/write locks)
+    - New optimized Binomial and Multinomial samplers.
+    - Parallelize some additional loops, and tune rayon min lengths for better cpu utilization.
+    - Use ahash in place of HashMap and BTreeMap in some places.
+    - Many other minor optimizations.
+
+# 3.0.12
+  - Fix numerical error crash occurring in some Visium HD data.
+  - Fix an incompatibility with spatialdata zarr output and newer versions of the python package.
+
+# 3.0.11
+
+  - Minor optimizations to zarr output size.
+  - Exclude unexpressed genes with visium or anndata input.
+  - Don't track count credible intervals, reducing memory usage.
+  - Support zarr-encoded AnnData input with the `--anndata` argument.
+  - Add an `--unmodeled-fixed-cells` argument to exclude the fixed cells from the background model.
+
+# 3.0.10
+
+ - Fix compile errors due to breaking changes in zarrs.
+
+# 3.0.9
+ - Add a `--cellpose-masks-fixed` to allow a second auxiliary segmentation mask to include
+   additional prior cell boundaries that will not be modified by proseg.
+ - Support reading transcripts and cell polygon from spatialdata zarr files.
+ - Fix some cases missed by `--enforce-connectivity`.
+
+# 3.0.8
+ - Fix an issue where `original_cell_id` was incorrectly assigned for some data.
+
+# 3.0.7
+ - Reworks the prior on cell compactness to fix excessive expansion of empty or near-empty cells.
+ - Fix a bug that made transcript repositioning on the z-axis very unlikely.
+
+# 3.0.6
+ - Fix an error generating transcript metadata when run using `--cellpose-masks`
+
+# 3.0.5
+ - Fix an occasional error where the last original cell id was missing causing cell metadata output to panic.
+
+# 3.0.4
+ - Fix an error keeping track of original transcript ids which can cause
+   incorrect counts when converted with `xeniumranger import-segmentation`.
+
+# 3.0.3
+ - Change spatialdata output to use consensus 2d polygons, rather than union polygons
+   which tend to exagerrate the degree of overlap.
+
+# 3.0.2
+ - Fix `proseg-to-baysor` not correctly marking transcripts as noise resulting in an inconsistent
+   count matrix after importing with xenium ranger.
+ - Fix an error when `--check-consistency` is used.
+ - Fix a minor inconsistency between the count matrix and transcript metadata
+   caused by them being out of sync by one sample.
+
+# 3.0.1
+ - Fix an error on xenium data where cells crossing fov boundaries were split.
+
+# 3.0.0
+  This is a major revision which has some potentially breaking changes. See
+  the "Migrating to Proseg 3" section in the README.md file.
+
+  - Support for Visium HD
+  - Extensive optimization work (runs should significantly less memory and be up to twice as fast now).
+  - Outputting zarr formatted spatialdata objects
+  - Outputting sparse matrices in mtx format
+  - Ability to initialize directly from cell masks from Cellpose and similar image based segmentation methods
+  - Revisions to the noise model to capture the dependency on local transcript density
+
+# 2.0.6
+  - Fix docker image for some nextflow configurations.
+  - Fix negative controls included in newer CosMx data.
+
+# 2.0.5
+  - Fix potentially incorrect `original_cell_id` in cell metadata.
+
+# 2.0.4
+  - Fix potential empty cells when large number of voxel layers are used.
+  - Remove `--double-z-layers` which was always true, and add `--no-z-layer-doubling` to turn it off.
+  - Use a default min qv of 20 in Xenium.
+
+# 2.0.3
+  - Fix crash on data with fewer than 300 genes.
+  - Output corresponding original cell ids to the cell metadata table.
+
+# 2.0.2
+  - Updating dependencies to avoid some compilation errors.
+  - Optimizations (should be 20-30% faster on high-plex data)
+  - Automatic untilting of samples that are tilted on the z-axis.
+
+# 2.0.1
+  - Implement a partial factorization model to avoid model failures on high-plex
+    data like Xenium Prime.
+  - Fix `--initial-voxel-size` not actually doing anything.
+  - Include an experiment `--visiumhd` preset, that has had to be tested.
+
+# 1.1.9
+  - Fix potential error using rust >=1.81 (Issue #41)
+
+# 1.1.8
+  - Fix error reading some newer Xenium parquet files.
+
+# 1.1.7
+  - Fix error compiling proseg using rust 1.81.
+
+# 1.1.6
+  - Fix errors when importing segmentation using xeniumranger 3. (Issue #38)
+
+# 1.1.5
+  - Fix failure to read certain .csv.gz input files. (Issue #29)
+
+# 1.1.4
+  - Fix error reading csv transcript metadata in `proseg-to-baysor`
+
+# 1.1.3
+  - Fix some cases in which weird or invalid polygons are reported.
+
+# 1.1.2
+  - Use newer Xenium parquet format.
+
+# 1.1.1
+  - Support for running on Xenium parquet files.
+  - Switch from unmaintained arrow2 crate to maintained arrow/parquet crates.
+
+# 1.1.0
+
+  - `--output-cell-polygons` will now generate non-overlapping "consensus"
+    polygons formed by taking the dominant cell assignment at each x/y location.
+    The old behavior, taking the union along the z-axis, can be generated now with
+    `--output-union-cell-polygons`.
+  - Default behavior is now to initialize voxels in 1D and double z-resolution twice,
+    which seems to lead to less spurious overlap between cells.
+  - `proseg-to-baysor` now filter out empty cells that were causing issues with
+    Xenium Ranger.
+
+# 1.0.6
+
+  - Add an option `--use-cell-initialization` to ignore compartment information
+    and initialize with cell assignment.
+  - Change the behavior of the `--cosmx` preset to support newer CosMx
+    transcript tables.
+
+# 1.0.5
+
+  - Fix overflow error generating transcript metadata table with large number
+    of transcripts. (Fixes #12)
+
+# 1.0.4
+
+  - Include fov in transcript metadata and cell metadata (Issue #6)
+  - Read fov column from more recent Xenium data
+  - Remove polygon interiors that form when flattening polygons into 2D.
+
+# 1.0.3
+
+  - Include a `--merscope` preset to work with MERSCOPE `detected_transcripts.csv.gz`
+    files. Mark the `--merfish` preset as deprecated. (Issue #5)
+
+# 1.0.2
+
+  - Fix overflow in `proseg-to-baysor` with large number of cells.
+
+# 1.0.1
+
+  - Fix for incompatibility with different CosMx file versions (Issue #3)
+  - Fix incompatibility with xenium ranger 2.0 (Issue #2)
+
+# 1.0.0
+
+Initial release