Skip to contents

Changelog

Source: NEWS.md

LIANA 0.1.12 (24.02.22)

  • Added rank_aggregate to generate both specificity and magnitude rank aggregates. Essentially runs liana_aggregate twice with different aggregate_how parameters and joins.
  • Added invert_specificity, invert_magnitude, invert_function parameters to liana_dotplot.
  • Added decompose_tensor as a function to run only the decomposition on a pre-built Tensor.
  • Aggregation can now be performed also via liana_bysample, takes aggregate_how parameter, which allows magnitude, specificity, or both.
  • Added preprocess_scores function that handles the conversion of liana’s scores to Tensor format.
  • Added additional tests related to Tensor-cell2cell

LIANA 0.1.11 (06.02.22)

  • Fixed issue with sample_col in liana_tensor functions
  • Removed redundant scconnect code
  • Merged #89 to fix typo in liana_aggregate documentation
  • Fix bug with column duplicates in cellchat_formatDB.
  • passing “all” to select_resource will now only return human resources.
  • add minor condition improvements to liana_wrap PR#92

LIANA 0.1.10 (23.01.22)

  • Fixed issues in the generate_lr_geneset function, and other typos and by mistake hardcoded values.
  • Bumped up the version of Tensor-cell2cell to the latest one, and it will now return error values that could be used to estimate the elbow curve.
  • Added an example of the elbow plot used to guide the number of ranks to be considered.
  • Extended options by which the cell2cell module can be loaded.
  • Fixed an issue where return_all was being passed to external methods
  • Added an example elbow curve in the tensor tutorial.
  • LIANA’s doplot will now keep the order of the original dataframe passed to it.

LIANA 0.1.9 (13.12.22)

Changes

  • return_all parameter was included to liana_wrap. return_all enables the return of all interactions by liana, not only the ones that pass the expr_prop threshold. Those that don’t pass the threshold are assigned the worst possible score, and a lrs.to.keep flag that indicates whether the interaction passed the threshold.

  • supp_columns was included to liana_wrap, which allows additional columns to be added to the output of any method.

Minor changes

  • Min prop for complexes is now explicitly assigned to the minimum expr. prop. across all subunits This deals with edge cases with non-expression ligand/receptor scores (e.g. z-score) where the lower score subunit is the one with higher expression proportion. Hence, this is intended to make all methods consistent according to which interactions are returned, regardless of which subunit has the lower score.
  • aggregate_how parameter added to liana_aggregate to allow the aggregation by specificity and magnitude scores.

LIANA 0.1.8 (08.11.22)

New Implementations

  • Untargeted between-condition (context/sample) decomposition of cell-cell communication latent patterns /w tensor_cell2cell. Makes use of basilisk to automatically set-up a conda env for liana.
  • added min_cells parameter to liana_wrap, to exclude any cell identity which does not pass a minimum cells threshold.

Changes

  • Mouse Consensus resource is now provided by default.
  • The intracellular OmniPath vignette was removed. An updated and more user-friendly one will be provided in next updates. In the meantime, the old one can still be downloaded from drive
  • Source and Target titles are now plotted by the liana_dotplot
  • added explicit error if idents_col was not found in metadata/colData

LIANA 0.1.7 (13.10.22)

Changes

  • Changed the way ties are handles in liana_aggregate. Namely, I previously assigned the minimum rank, but this resulted in ties getting lower p-values than they should, particularly for scores with a lot of ties (e.g. CPDB p-value).

  • Fixed an issue where some subunits of 0 expr_prop would not get filtered. This was observed due to previous changes to .filt_liana_pipe in 0.1.6, where some subunits were filtered before recomplexifying.

  • Fixed an issue in which some NATMI complexes would be missing due to recomplexify being done on both .expr and .sum columns. These are now seperated into columns, which are the ones for which I account for complexes, and add_cols, the ones that are additional - no need to account for complexes (e.g. also global_mean).

Minor Changes

  • I now refer to SCPubr in liana’s tutorial.
  • Throw exception for NAs in cell idents ` Remove duplicated rows from orthologous resource

Minor changes

  • I now refer to SCpubr in the tutorial for more advanced plots.

LIANA 0.1.6 (11.08.22)

Changes

  • Fixed an issue where interactions with complexes will not filtered be according to expr_prop for some methods. I now filter twice - once via .filt_liana_pipe for computational speed, and once after recomplexify to also remove the complexes with expr_prop <= X. Will now also filter crosstalk_scores to expr_prop.

  • In the edgecase of complexes with subunits with equal expression, LIANA’s internal methods will not arbitrarily discard duplicate complex interactions.

  • Will now return expr_prop for each method. Note that this information is discarded by liana_aggregate.

  • liana_doplot function is now more explicit in the way interactions are selected. Will now take topn and return the highest ranked interactions. Size of dots is also more distinguishable by default and the user can now pass a customizable value for the size range.

  • Added a rank_method helper function to rank single methods according to specificity and magnitude.

  • Removed ~20 bad quality interactions from the Consensus resource.

  • Minor changes on filtering SCE object in liana_pipe to ensure all complex subunits are present in the sce

LIANA 0.1.5 (04.07.22)

Changes

  • Re-implemented the RRA method from Kolde et al., 2012, as a consequence of the removal of the RobustRankAggregate package from CRAN.

  • Integrate generate_homologs with OmniPath’s homologene database. This allows homology conversion by simply passing an organism ID. Also, handles complicated cases, such as complex subunits with one-to-many mapping homologs.

LIANA 0.1.4 (18.06.22)

Changes

  • Add prod_weight to NATMI’s score. This is the weight that both Connectome and NATMI suggest for between-condition comparisons. Add NATMI to the housekeeping aggregate ranking.

  • Enable weighing of interactions by cell pairs (using a DF in which each cell pair has an assigned weight). This would typically be done by spatial constraints, etc. These weights can also be used to mask any cell-pair interactions which are not relevant (by assigning weights of 0). This currently assumes that the weights would be between 0 to 1 - to be extended. Tutorial on this /w appropriate spatial weight generation to be written.

Minor Changes

  • By default, the base for logFC will now be automatically assigned depending on the object passed to LIANA, i.e. .antilog1m for SCE will use 2 as base, and Euler’s number for Seurat. One could also pass the base they wish to use via liana_wrap.
  • Automate website deployment to gh-pages and run R checks on push.

LIANA 0.1.3 (15.05.22)

Changes

  • Changed the aggregation columns of liana_aggregate, as in some cases methods would assign different subunits as the minimum, which results in redundancies for the same complex. As such, liana_aggregate will now return only the complex columns, nevertheless, the methods will still return both the minimum (lowest expressed subunit) and it’s corresponding complex.

  • base used to calculate logFC (via get_log2FC) can now be passed as a parameter to liana_wrap via liana_pipe.params. Passing NaN to base would result in log2FC calculation using the raw counts without any pre-processing (e.g. no batch correction, etc).

The base is by default set to 2, assuming that log2 transformation is performed following library size normalization, and thus preserving the normalization, while reverting back to ~counts.

Minor Changes

  • Extended the heatmap to allow filtering down to certain cell types.
  • Removed redundant/leftover code

New implementations

  • A chord plot for interaction frequncies was included.

LIANA 0.1.2 (03.05.22)

New Implementations

  • Frequency Heatmap available via the heat_freq functions, added due to being common requests. This heatmap was inspired by CellPhoneDB and CellChat.

Changes

  • Extended basic tutorial to accommodate new heatmap plots.

Minor changes

  • Allow labels to be passed to liana_dotplot
  • Cleaned up docs, dependencies, examples, and warnings

LIANA 0.1.1 (26.04.22)

Changes

  • Change the order of non-expressed genes and empty droplet filtering. I now appropriately filter cells in the sce object after limiting the gene universe to ligands and receptors in the resource.

Minor changes

  • Appropriately pass verbose to .filter_sce

  • Silence expected warning in cellchat_formatDB

LIANA 0.1.0 (20.04.22)

Changes

  • For logFC_mean, rather than normalizing the counts by library size, I instead inverse log the counts and use those to calculate log2FC. This is to preserve any prior correction of the counts, i.e. mainly for consistency with the rest of the methods.

Minor Changes

  • Flipped x and y axes dotplot according to feedback.

  • Added .default_fun parameter to generate_orthologues

  • Minor code clean up

  • Added examples to main exported functions docs

  • Removed several low quality interactions from the Consensus resource

  • CellChat will now work with the simplified format of intercell (i.e. Consensus resource).

  • CellChat and Squidpy should now be called using call_cellchat and call_squidpy, respectively.

LIANA 0.0.9 (23.03.2022)

New Implementations

  • We now provide a tutorial dedicated to orthology conversion of the resources in LIANA

Minor Changes

LIANA will now check if: - There is enough of a gene intesect between the resource and the data (i.e. if they are both for human)

  • There are negative counts

  • An assay with normalized counts was been provided (i.e. if the data/logcounts slot is scaled).

  • LIANA will now convert non-sparse matrices to sparse.

LIANA 0.0.8 (05.03.2022)

Changes

  • Reduced dependencies (specifically Seurat and OmniPathR)

Minor Changes

  • testthat tests external methods only if requested explicitly
  • Readme updated - clarified and accordingly describes the Consensus resource as default

New Features

  • idents_col is can now be explicitly passed to liana_wrap, if not provided defaults to the active idents/colLabels for SCE and Seurat, respectively.
  • verbose param allows to omit any messages and warnings from LIANA
  • assay can now be passed explicitly when working with a Seurat object, defaults to the active one otherwise

LIANA 0.0.7

Changes

  • LIANA will now use the Consensus resource by default. This is a highly-literature supported resource, generated using similar filtering steps as the ‘OmniPath’ (old default) resource. This resource is similar in size (~4,700 interactions), but contains a higher complex and curation content.

  • All resources might show some very minor changes related to an update of UniProt IDs and homology-conversion improvements.

  • LIANA now uses mean0 to account for heteromeric complexes, i.e. the mean is computed, unless there is a value of 0, then 0 is returned. This means that any complex, the subunit of which is not expressed is filtered. LIANA now also appropriately accepts any custom function to account for complexes.

  • liana_aggregate now groups by ligand.complex and receptor.complex as well as the subunits, and hence returns a all of those columns

Minor Changes

  • Added option to show complexes on dotplot and is now the default option

  • Documentation improvements

  • decomplexify function is now exported

  • liana_aggregate will no longer return a median_rank, it’s largely redundant.

  • re-arranged the column order of liana_aggregate due to the addition of .complex columns

  • Replaced min0 (used to obtain closest to 0 value) to min -> relevant for z-scores used in Connectome.

Bugs

  • Complexes with missing subunits are not correctly assigned as ‘missing’ and hence filtered/treated as non-expressed.

  • Fixed a bug where LIANA will return the minimum subunit expression, instead of the mean for some methods. This stemmed from not properly passing the incorrect complex_policy to certain methods, i.e. they were getting a hard-coded value instead.

  • Remove decomplexify logical from liana_call and liana_pipe -> redundant.

  • edge case fix: liana_aggregate should now rank interactions with the same subunits, but coming from different complexes seperately

LIANA 0.0.6

New implementations

  • LIANA has now been optimized in terms of RAM, by swapping all internal function to rely solely on the BioConductor single-cell framework (for all internal methods).

  • LIANA now accepts both SingleCellExperiment and Seurat objects as input.

  • added liana_dotplot as a basic, but flexible, dotplot function for LIANA output. (+ tests)

Changes

  • LIANA will now perform a basic filtering step, where all genes and cells with 0 summed counts are removed.

  • global_mean is now calculated in a more efficient manner.

  • assay.type in liana_pipe was passed to get_logFC would result in using the logcounts, rather than the library-normalized counts.

Bug Fixes

  • Fixed a bug where incorrectly passing method names in different cases results in an error.

Deprecated

  • External LIANA methods (i.e. call_) are now deprecated. The pipelines will be maintained solely for power users, who intend to benchmark the original implementations, but will not be the focus of any downstream analyses. These will be solely developed for the internal (or re-implemented methods). These still rely on a SeuratObject as interface, but will now accept both sce and seurat as input.

LIANA 0.0.5

Changes

  • I now filter the Crosstalk scores to include only those > 0. Otherwise, LIANA would return all possible combinations of clusters and interactions, which would be simply NAs and 0s for Crosstalk scores. Should do the same for Connectome (>0).

  • CellChat and Crosstalk scores/cytotalk will no longer by called by default by liana_wrap. However, it both are available as an option to be passed to the method parameter.

  • I now filter all methods by expr_prop. This is done in a slightly different manner for Connectome’s scaled weights and crosstalk scores, since they require all pairs/clusters to be present to appropriately calculate their scores. Thus, for them we filter after we calculate the scores, while for the others methods we pre-filter.

  • We now provide a tutorial how to make use of intracellular OmniPath as well as how to combine LIANA with NicheNet

LIANA 0.0.4

  • The OmniPath resource had a major update.

  • CellCall and Cellinker resources were added, while talklr was removed. The OmniPath resources itself was filtered further and 1,000 lower quality interactions were excluded. Further improvements were made to all resources, most of which were minor. Changes worth mentioning were made to ICELLNET (updated to latest resource version), CellPhoneDB (was filtered for ambigous interactions), and CellChatDB was filtered for mislaballed interactions.

LIANA 0.0.3

Improvements

  • The R re-implementation of CellPhoneDBv2’s permutation algorithm was optimized to work with sparse matrices (and is now uqicker), and set as the default option in LIANA (replacing the re-implementation of the same algorithm from squidpy)

  • Custom proportion filtering - Connectome and CytoTalk are now not filtered by expr_prop as this affects the way that their scores are calculated, since they require all clusters/cluster pairs to be present to appropriately scale or normalize their scores.

Bug Fixes

  • Fixed an issue where logFC was assigned only the value of the ligand

LIANA 0.0.2

New Features

  • Crosstalk scores inspired by Cytotalk were added. In contrast to the CytoTalk, in our calculation CTS with ligand or receptor with PEM of 0 are assigned 0 CTS. Furthermore, we use the inverse of the non-self-talk scores calculated in CytoTalk to also allow for autocrine signalling interactions, and thus make Crosstalk scores comparable to the rest of the methods in LIANA. Finally, as part of LIANA, CytoTalk’s re-implemented scores would not take account of complexes and we also apply liana-specifc filtering such as according to expr_prop. Worth noting, we only re-implement the cross-talk scores, but we don’t include the intracellular part of Cytotalk.

Changes

  • Changed expr_thresh to 0.1, based on lack of improvement in performance when using 0.2, hence opted out for the less conservative threshold as default
  • Changed the way that default parameters are passed to each method
  • Enabled housekeeping score aggregation for external methods (needed for revisions) via .score_housekeep
  • Fixed Bug where external methods could not be called with their default DB. The resource is now always decomplexified
  • Seurat Testdata is now properly normalized
  • liana_aggragate will now by defaul dissociate complexe for CellChat Complexes
  • Added tests for changes

LIANA 0.0.1

New Features

liana_wrap and liana_aggragate as the two highest level functions to run all the methods in liana and aggragate them, respectively.

Re-implemented the following scores in LIANA:

  • logFC
  • NATMI specificity edges
  • Connectome scaled_weights
  • CellPhoneDB algorithm
  • SingleCellSignalR LRScore

each called via liana_call, which leverages the statistics provided by liana_pipe,

Others

  • Not re-implemented method score names now start with call_*

  • decomplexify and recomplexify as functions used to dissociate complexes in resources and account for complexes of the re-implemented methods above

  • liana_aggragate - a handy wrapper to aggregate results

  • LIANA and LIANA++ are now the user-friendly and benchmark version of LIANA, respectively

  • A webpage with vignettes showing the validity of the re-implemented methods, a developer/benchmark-focused vignette, and a vignette to customize OmniPath

Bug fixes

A number of fixes were implemented thanks to the early stage users. Thanks you.