CoChat

LIANA+ provides an all-in-one framework for cell–cell communication inference

Open access
Dimitrov, Daniel, Schäfer, Philipp Sven Lars, Farr, Elias, Rodriguez-Mier, Pablo, Lobentanzer, Sebastian, Badia-i-Mompel, Pau, Dugourd, Aurelien, Tanevski, Jovan, Ramirez Flores, Ricardo Omar, Saez-Rodriguez, Julio · 2024 · Nature Cell Biology

The DOI matches its Crossref record; 118 of the 118 references it cites were found in a registry or an index.

Abstract

The growing availability of single-cell and spatially resolved transcriptomics has led to the development of many approaches to infer cell–cell communication, each capturing only a partial view of the complex landscape of intercellular signalling. Here we present LIANA+, a scalable framework built around a rich knowledge base to decode coordinated inter- and intracellular signalling events from single- and multi-condition datasets in both single-cell and spatially resolved data. By extending and unifying established methodologies, LIANA+ provides a comprehensive set of synergistic components to study cell–cell communication via diverse molecular mediators, including those measured in multi-omics data. LIANA+ is accessible at https://github.com/saezlab/liana-py with extensive vignettes ( https://liana-py.readthedocs.io/ ) and provides an all-in-one solution to intercellular communication inference.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41556-024-01469-w

Cites

118 of 118 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 LIANA+ framework overview. a, LIANA+ re-implements and adapts eight ligand–receptor methods to infer interactions from single-cell data, along with a flexible consensus that can integrate any combination of these methods. b, LIANA+ implements multi-view learning as well as eight local metrics to respectively capture global and local interactions from spatially resolved omics data. c, LIANA+ includes diverse strategies to identify deregulated CCC events across conditions: (1) differential CCC analysis with PyDESeq2 (refs. 47,48) for hypothesis-driven exploration and (2) unsupervised (hypothesis-free) approaches including standard NMF or higher-order factorizations via Tensor-cell2cell20 and multi-view factor analysis43. d, LIANA+ connects intercellular interactions to intracellular signalling pathways using sign-coherent network optimization. e, LIANA+ is built on a rich knowledge base—OmniPath22 and BioCypher59—which comprise ligand–receptor interactions and annotations, including such mediated by metabolites68, as well as intracellular knowledge, such as signalling pathways and TFs. Finally, all components (a–e) of LIANA+ are applicable to both dissociated single-cell and spatially resolved multi-omics data. aFor spatially weighted Spearman correlation, along with the standard metric, we implemented its masked version from scHOT32; for Moran’s R we adapted both the global and local versions from SpatialDM8 and for the spatially weighted product, we also included a max-normalized version.
Fig. 1 LIANA+ framework overview. a, LIANA+ re-implements and adapts eight ligand–receptor methods to infer interactions from single-cell data, along with a flexible consensus that can integrate any combination of these methods. b, LIANA+ implements multi-view learning as well as eight local metrics to respectively capture global and local interactions from spatially resolved omics data. c, LIANA+ includes diverse strategies to identify deregulated CCC events across conditions: (1) differential CCC analysis with PyDESeq2 (refs. 47,48) for hypothesis-driven exploration and (2) unsupervised (hypothesis-free) approaches including standard NMF or higher-order factorizations via Tensor-cell2cell20 and multi-view factor analysis43. d, LIANA+ connects intercellular interactions to intracellular signalling pathways using sign-coherent network optimization. e, LIANA+ is built on a rich knowledge base—OmniPath22 and BioCypher59—which comprise ligand–receptor interactions and annotations, including such mediated by metabolites68, as well as intracellular knowledge, such as signalling pathways and TFs. Finally, all components (a–e) of LIANA+ are applicable to both dissociated single-cell and spatially resolved multi-omics data. aFor spatially weighted Spearman correlation, along with the standard metric, we implemented its masked version from scHOT32; for Moran’s R we adapted both the global and local versions from SpatialDM8 and for the spatially weighted product, we also included a max-normalized version.CC BY · PubMed Central
Fig. 2 LIANA+ uses standardized inputs and outputs to streamline the inference intercellular and intracellular signalling. a, LIANA+ accepts unimodal (AnnData) or multi-modal (MuData) data objects as inputs with (optional) prior knowledge and/or spatial information. These are then transformed into data frames with aggregated interaction results or unimodal objects with statistics at the individual spot or cell level. To enable the inference of CCC across modalities, the methods implemented in LIANA+ accept MuData objects77 as input. These provide essential functionalities to load and store multi-modal data77, and can be thought of as an extension of AnnData objects78, which are the default input of LIANA+ when working with unimodal single-cell or spatial data. b,c, LIANA+ makes use of existing prior knowledge frameworks (b)22,59,68 to annotate interactions according to, for example, pathways, disease or location (c). d, Similarly, it uses this prior knowledge to infer putative causal (sign-coherent) signalling networks49,79, emanating from ligand–receptor interactions down to active TFs. e, The TF activities can be estimated by making use of generalistic regulon prior knowledge80 and standard enrichment analyses81.
Fig. 2 LIANA+ uses standardized inputs and outputs to streamline the inference intercellular and intracellular signalling. a, LIANA+ accepts unimodal (AnnData) or multi-modal (MuData) data objects as inputs with (optional) prior knowledge and/or spatial information. These are then transformed into data frames with aggregated interaction results or unimodal objects with statistics at the individual spot or cell level. To enable the inference of CCC across modalities, the methods implemented in LIANA+ accept MuData objects77 as input. These provide essential functionalities to load and store multi-modal data77, and can be thought of as an extension of AnnData objects78, which are the default input of LIANA+ when working with unimodal single-cell or spatial data. b,c, LIANA+ makes use of existing prior knowledge frameworks (b)22,59,68 to annotate interactions according to, for example, pathways, disease or location (c). d, Similarly, it uses this prior knowledge to infer putative causal (sign-coherent) signalling networks49,79, emanating from ligand–receptor interactions down to active TFs. e, The TF activities can be estimated by making use of generalistic regulon prior knowledge80 and standard enrichment analyses81.CC BY · PubMed Central
Fig. 3 LIANA+ models intercellular communication from spatial multi-omics data. a, Spatially resolved transcriptomics (10X Visium) and metabolomics (matrix-assisted laser desorption/ionization mass spectrometry imaging) yield two matrices with different sets of features for each modality (f1 or f2), and observations (n or m) which correspond to different locations captured on the same tissue section26. b, Parkinson’s disease mouse model annotated for striatum in intact and lesioned hemispheres, with three replicates (3✕). c, Multi-view modelling integrates metabolite peak intensities, brain-specific receptor expression and cell-type proportions to identify their spatial relationships. This approach enables the estimation of joint performance and individual contributions of receptor expression and cell-type proportions in predicting metabolite peak intensities. Cell-type proportions were deconvoluted35 using a murine single-cell atlas36 as a reference. d, Dopamine predictors ranked according to their median importance (y axis; ordinary least squares t-values), with names shown for the top three predictors: Drd2 and MSNs 1 and 2. e, Normalized dopamine peak intensities. f, log1p Drd2 receptor gene expression. g–i, Local interactions between dopamine and its canonical D2R receptor, encoded by Drd2, as measured by spatially weighted cosine similarity (g), its corresponding uncorrected permutation P values (h) and interaction categories (i). The images showcase slide B1 from experiment V11L12-109 (ref. 36). Source numerical data are available in source data. Source data
Fig. 3 LIANA+ models intercellular communication from spatial multi-omics data. a, Spatially resolved transcriptomics (10X Visium) and metabolomics (matrix-assisted laser desorption/ionization mass spectrometry imaging) yield two matrices with different sets of features for each modality (f1 or f2), and observations (n or m) which correspond to different locations captured on the same tissue section26. b, Parkinson’s disease mouse model annotated for striatum in intact and lesioned hemispheres, with three replicates (3✕). c, Multi-view modelling integrates metabolite peak intensities, brain-specific receptor expression and cell-type proportions to identify their spatial relationships. This approach enables the estimation of joint performance and individual contributions of receptor expression and cell-type proportions in predicting metabolite peak intensities. Cell-type proportions were deconvoluted35 using a murine single-cell atlas36 as a reference. d, Dopamine predictors ranked according to their median importance (y axis; ordinary least squares t-values), with names shown for the top three predictors: Drd2 and MSNs 1 and 2. e, Normalized dopamine peak intensities. f, log1p Drd2 receptor gene expression. g–i, Local interactions between dopamine and its canonical D2R receptor, encoded by Drd2, as measured by spatially weighted cosine similarity (g), its corresponding uncorrected permutation P values (h) and interaction categories (i). The images showcase slide B1 from experiment V11L12-109 (ref. 36). Source numerical data are available in source data. Source dataCC BY · PubMed Central
Fig. 4 Comparison of ligand–receptor inference methods using the spatial co-localization of cell types and ligand–receptors as assumed truth. Each method, as re-implemented in LIANA+, and its individual ligand-receptor (LR) scoring functions are represented by a different box colour. Scoring functions that reflect the magnitude strength of an interaction are shown in dark red, while those that capture cell-type specificity in black. a, Quantification of the performance of each method’s scoring functions using the AUROC. b, The balanced accuracy for each of the methods filtered according to their suggested false positive filtering thresholds. c,d, The normalized F1 scores following filtering for each score (*) (c) and method (d). For each metric, a score of 0.5, denoted by the dashed red line, indicates random performance. Note that while we show 0 to 1 for normalized F1, unlike balanced accuracy and AUROC, it is bound between negative and positive infinity (Methods). Both NATMI and Connectome use expression products (Product*) as a measure of magnitude strength. The boxes were ordered according to the median performance across datasets (central line in cyan within each box; n = 5 datasets). The box hinges represent the first and third quartiles, and the whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Outliers are depicted as individual hollow points beyond the whiskers. Source numerical data are available in source data. Source data
Fig. 4 Comparison of ligand–receptor inference methods using the spatial co-localization of cell types and ligand–receptors as assumed truth. Each method, as re-implemented in LIANA+, and its individual ligand-receptor (LR) scoring functions are represented by a different box colour. Scoring functions that reflect the magnitude strength of an interaction are shown in dark red, while those that capture cell-type specificity in black. a, Quantification of the performance of each method’s scoring functions using the AUROC. b, The balanced accuracy for each of the methods filtered according to their suggested false positive filtering thresholds. c,d, The normalized F1 scores following filtering for each score (*) (c) and method (d). For each metric, a score of 0.5, denoted by the dashed red line, indicates random performance. Note that while we show 0 to 1 for normalized F1, unlike balanced accuracy and AUROC, it is bound between negative and positive infinity (Methods). Both NATMI and Connectome use expression products (Product*) as a measure of magnitude strength. The boxes were ordered according to the median performance across datasets (central line in cyan within each box; n = 5 datasets). The box hinges represent the first and third quartiles, and the whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Outliers are depicted as individual hollow points beyond the whiskers. Source numerical data are available in source data. Source dataCC BY · PubMed Central
Fig. 5 LIANA+ models intercellular communication from dissociated and spatially resolved transcriptomics data. a, Sampling sites from human myocardial infarction patients. b, Overview of NMF applied to local interactions inferred for each location/observation across all slides. c, log2-transformed odds ratios and Fisher’s exact test P values representing the enrichment or depletion of fibrotic, ischaemic and myogenic labels in each of the factors inferred by NMF. d, Pathway enrichment50 of NMF ligand–receptor loadings. Asterisks indicate FDR-corrected P values <0.05, with the colour map distinguishing between positive (red) and negative (blue) enrichment. e, NMF factor 1 scores per observation in an ischaemic tissue section. f,g, Spatially weighted cosine similarity of FN1 (f) and SPP1 (g) with the ITGA5 and ITGB1 complex, respectively, in a selected (ACH0014) sample. h, The procedure to decompose ligand–receptor interactions inferred between cell types from dissociated single-nucleus data samples into factors and corresponding feature sets. i, Multi-view factor scores following ligand–receptor score decomposition with one-way analysis of variance P value across ischaemic, fibrotic and myogenic samples (n = 29). The central line within each box marks the median, with the box hinges representing the first and third quartiles. The whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Outliers are depicted as individual hollow points. j, Multi-view factorization factor 1 variance explained across cell-type pairs (views). Abbreviations include cardiomyocytes (CM), endothelial cells (EN), FB, MY, pericytes (PC) and vascular smooth muscle cells (VM). k, To link deregulated inter- and intracellular signalling events, we combine ligand–receptor prior knowledge with differential contrast statistics between ischaemic samples versus the rest. Then, using knowledge of intracellular protein–protein interactions and TF regulons, we identify sign-coherent subnetworks that connect intercellular interactions (start nodes) with deregulated TFs (end nodes; Methods). l, A subset of interactions, the ligand and/or receptors of which are known to play a role in fibrosis27,53,55 and were deregulated in FB and/or MY types. Only interactions with the highest loadings (>95th percentile) from the NMF analysis on spatially informed local ligand–receptor interactions were included in the differential expression analysis. Asterisks signify genes with FDR <0.05, and the colour bar corresponds to log2FC47,48. m, Sign-coherent signalling network originating from FN1 and SPP1 and propagating down to TFs deregulated in MY in ischaemia. n, Corresponding legend for the network in m. Source numerical data are available in source data. Source data
Fig. 5 LIANA+ models intercellular communication from dissociated and spatially resolved transcriptomics data. a, Sampling sites from human myocardial infarction patients. b, Overview of NMF applied to local interactions inferred for each location/observation across all slides. c, log2-transformed odds ratios and Fisher’s exact test P values representing the enrichment or depletion of fibrotic, ischaemic and myogenic labels in each of the factors inferred by NMF. d, Pathway enrichment50 of NMF ligand–receptor loadings. Asterisks indicate FDR-corrected P values <0.05, with the colour map distinguishing between positive (red) and negative (blue) enrichment. e, NMF factor 1 scores per observation in an ischaemic tissue section. f,g, Spatially weighted cosine similarity of FN1 (f) and SPP1 (g) with the ITGA5 and ITGB1 complex, respectively, in a selected (ACH0014) sample. h, The procedure to decompose ligand–receptor interactions inferred between cell types from dissociated single-nucleus data samples into factors and corresponding feature sets. i, Multi-view factor scores following ligand–receptor score decomposition with one-way analysis of variance P value across ischaemic, fibrotic and myogenic samples (n = 29). The central line within each box marks the median, with the box hinges representing the first and third quartiles. The whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Outliers are depicted as individual hollow points. j, Multi-view factorization factor 1 variance explained across cell-type pairs (views). Abbreviations include cardiomyocytes (CM), endothelial cells (EN), FB, MY, pericytes (PC) and vascular smooth muscle cells (VM). k, To link deregulated inter- and intracellular signalling events, we combine ligand–receptor prior knowledge with differential contrast statistics between ischaemic samples versus the rest. Then, using knowledge of intracellular protein–protein interactions and TF regulons, we identify sign-coherent subnetworks that connect intercellular interactions (start nodes) with deregulated TFs (end nodes; Methods). l, A subset of interactions, the ligand and/or receptors of which are known to play a role in fibrosis27,53,55 and were deregulated in FB and/or MY types. Only interactions with the highest loadings (>95th percentile) from the NMF analysis on spatially informed local ligand–receptor interactions were included in the differential expression analysis. Asterisks signify genes with FDR <0.05, and the colour bar corresponds to log2FC47,48. m, Sign-coherent signalling network originating from FN1 and SPP1 and propagating down to TFs deregulated in MY in ischaemia. n, Corresponding legend for the network in m. Source numerical data are available in source data. Source dataCC BY · PubMed Central
Extended Data Fig. 1 LIANA+ provides scalable implementations of cell–cell communication methods. Efficiency benchmark of A) dissociated methods and B) spatially weighted local metrics implemented in LIANA + . Time in log2 of seconds and peak total RAM (memory) usage are shown over a number of simulated datasets with observations (cells) ranging from 1000 to 100,000 (1k to 100k). Masked Spearman Correlation was excluded for the 100k dataset. C) Multi-view learning of cell-type spatial patterns across a range of views. RAM Usage is not shown as it was relatively constant at ~2,900MB ( ± 100MB) across views. D) Multi-view factor analysis used to identify intercellular programmes across a range of cell-type pairs (views). The median time in minutes and RAM usage are shown across five independent runs for each method. Source numerical data are available in source data. Source data
Extended Data Fig. 1 LIANA+ provides scalable implementations of cell–cell communication methods. Efficiency benchmark of A) dissociated methods and B) spatially weighted local metrics implemented in LIANA + . Time in log2 of seconds and peak total RAM (memory) usage are shown over a number of simulated datasets with observations (cells) ranging from 1000 to 100,000 (1k to 100k). Masked Spearman Correlation was excluded for the 100k dataset. C) Multi-view learning of cell-type spatial patterns across a range of views. RAM Usage is not shown as it was relatively constant at ~2,900MB ( ± 100MB) across views. D) Multi-view factor analysis used to identify intercellular programmes across a range of cell-type pairs (views). The median time in minutes and RAM usage are shown across five independent runs for each method. Source numerical data are available in source data. Source dataCC BY · PubMed Central
Extended Data Fig. 2 Spot-calling evaluations show that the local metrics in LIANA+ perform well at preserving biological information. A) AUROC and B) weighted F1 when using local metrics to classify malignant spots in breast cancer spatial transcriptomics data (n = 4 samples)76; C) R2 and D) RMSE when using local metrics to predict cell type proportions in heart spatial transcriptomics data27. The line in the boxplots represents the median with hinges showing the first and third quartiles and the whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Outliers are shown as diamonds. Source numerical data are available in source data. Source data
Extended Data Fig. 2 Spot-calling evaluations show that the local metrics in LIANA+ perform well at preserving biological information. A) AUROC and B) weighted F1 when using local metrics to classify malignant spots in breast cancer spatial transcriptomics data (n = 4 samples)76; C) R2 and D) RMSE when using local metrics to predict cell type proportions in heart spatial transcriptomics data27. The line in the boxplots represents the median with hinges showing the first and third quartiles and the whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Outliers are shown as diamonds. Source numerical data are available in source data. Source dataCC BY · PubMed Central
Extended Data Fig. 3 Multi-view modelling prediction performance per target in spatially resolved metabolite-transcriptome data. A) Top 10 metabolite peaks with the highest variance explained (R2), including Dopamine (in dark red) and 3-methoxytyramine (3-MT) peaks, as annotated in the original publication26. B) Relative contributions of views (receptors and cell types) when jointly predicting metabolite peak intensities. C) Contributions of each view in predicting dopamine. Error bars represent 95% confidence intervals around the mean. The displayed P-value was calculated using a two-sided paired t-test across replicates (n = 3). D-F) Differences between lesioned and intact hemispheres in Dopamine’s canonical Drd2 receptor, and medium MSN cell types 1 and 2. Uncorrected P-values were calculated using one-sided paired t-tests across replicates. The central line within each box marks the median, with the box hinges representing the first and third quartiles. The whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Source numerical data are available in source data. Source data
Extended Data Fig. 3 Multi-view modelling prediction performance per target in spatially resolved metabolite-transcriptome data. A) Top 10 metabolite peaks with the highest variance explained (R2), including Dopamine (in dark red) and 3-methoxytyramine (3-MT) peaks, as annotated in the original publication26. B) Relative contributions of views (receptors and cell types) when jointly predicting metabolite peak intensities. C) Contributions of each view in predicting dopamine. Error bars represent 95% confidence intervals around the mean. The displayed P-value was calculated using a two-sided paired t-test across replicates (n = 3). D-F) Differences between lesioned and intact hemispheres in Dopamine’s canonical Drd2 receptor, and medium MSN cell types 1 and 2. Uncorrected P-values were calculated using one-sided paired t-tests across replicates. The central line within each box marks the median, with the box hinges representing the first and third quartiles. The whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Source numerical data are available in source data. Source dataCC BY · PubMed Central
Extended Data Fig. 4 LIANA+ identifies local interactions from spatially resolved metabolite-transcriptome data. A) Lesioned and Intact hemispheres; B) Striatum annotation; C) Medium Spiny Neuron (MSN) 2 cell type proportions, and D) Drd1 expression. Interactions between Dopamine and MSNs E) 1 and F) 2, calculated using spatially-weighted cosine similarity. All panels show slide B1 from experiment V11L12-109.
Extended Data Fig. 4 LIANA+ identifies local interactions from spatially resolved metabolite-transcriptome data. A) Lesioned and Intact hemispheres; B) Striatum annotation; C) Medium Spiny Neuron (MSN) 2 cell type proportions, and D) Drd1 expression. Interactions between Dopamine and MSNs E) 1 and F) 2, calculated using spatially-weighted cosine similarity. All panels show slide B1 from experiment V11L12-109.CC BY · PubMed Central
Extended Data Fig. 5 LIANA+ captures intercellular programmes that separate patient samples according to different conditions. A) Classification setup to evaluate the ability of ligand-receptor methods, combined with Tensor-cell2cell and multi-view factor analysis (MOFA+), in separating conditions from multi-condition atlases (n = 5) in an unsupervised manner. B) Area under the receiver-operator curve (AUROC) and C) weighted F1 score, as calculated across 5 datasets, and the ‘Average’ performance. Source numerical data are available in source data. Source data
Extended Data Fig. 5 LIANA+ captures intercellular programmes that separate patient samples according to different conditions. A) Classification setup to evaluate the ability of ligand-receptor methods, combined with Tensor-cell2cell and multi-view factor analysis (MOFA+), in separating conditions from multi-condition atlases (n = 5) in an unsupervised manner. B) Area under the receiver-operator curve (AUROC) and C) weighted F1 score, as calculated across 5 datasets, and the ‘Average’ performance. Source numerical data are available in source data. Source dataCC BY · PubMed Central
Extended Data Fig. 6 Condition-relevant ligand-receptor interactions and spatially-co-localized cell types in Myocardial infarction. A) Top 30 interactions (with ‘FN1’, ‘TNC’, ‘THBS1’, and ‘SPP1 as ligands) in Factor 1 identified using NMF on local ligand-receptor metrics in spatially resolved 10X Visium heart samples. B) Importances (Median t-values clipped at 5 and -5; n = 28) from a spatially-weighted model predicting all possible cell type interactions from Myocardial infarction 10X Visium slides. Cell type interactions with Median t-value > 1.645 and R2 > 5% are marked with X. C) Multi-view Factorization feature (interaction) loadings following decomposition of ligand-receptor scores inferred from dissociated single-nucleus myocardial infarction data. The top 15 interactions with the highest interaction loadings are shown across all cell-type pairs. Abbreviations used include AD for Adipocytes, CM for Cardiomyocytes, EN for Endothelial cells, FB for Fibroblasts, PC for Pericytes, PR for Proliferating cells, VM for Vascular smooth muscle cells, NE for Neuronal cells, MY for Myeloid cells, MA for Mast cells, and LY for Lymphoid cells. Source numerical data are available in source data. Source data
Extended Data Fig. 6 Condition-relevant ligand-receptor interactions and spatially-co-localized cell types in Myocardial infarction. A) Top 30 interactions (with ‘FN1’, ‘TNC’, ‘THBS1’, and ‘SPP1 as ligands) in Factor 1 identified using NMF on local ligand-receptor metrics in spatially resolved 10X Visium heart samples. B) Importances (Median t-values clipped at 5 and -5; n = 28) from a spatially-weighted model predicting all possible cell type interactions from Myocardial infarction 10X Visium slides. Cell type interactions with Median t-value > 1.645 and R2 > 5% are marked with X. C) Multi-view Factorization feature (interaction) loadings following decomposition of ligand-receptor scores inferred from dissociated single-nucleus myocardial infarction data. The top 15 interactions with the highest interaction loadings are shown across all cell-type pairs. Abbreviations used include AD for Adipocytes, CM for Cardiomyocytes, EN for Endothelial cells, FB for Fibroblasts, PC for Pericytes, PR for Proliferating cells, VM for Vascular smooth muscle cells, NE for Neuronal cells, MY for Myeloid cells, MA for Mast cells, and LY for Lymphoid cells. Source numerical data are available in source data. Source dataCC BY · PubMed Central
Extended Data Fig. 7 LIANA+ generates interpretable and reliable predictions from complex datasets. A) Applying LIANA+ on Myocardial infarction substantially reduced the prediction space to yield an interpretable subset of predictions. (1) inference of the full possible prediction space of ligand-receptor interactions across 110 cell-type pairs, 28 samples, and 3 conditions. (2) Incorporation of spatially associated cell-type pairs (III). (3) Inclusion of information about spatially-colocalised ligand-receptor interactions (LRs), identified using (I) local scores with non-negative matrix factorization (II). (4) Multi-view factor analysis using LRs predicted from single-cell data and informed by the spatial analysis. (5) Targeted differential expression analysis to find significantly deregulated LRs in a specific pair of interacting literature-supported cell types (‡). (6) A causal network search to generate a specific mechanistic hypothesis linking LR interactions to downstream Transcription Factors (TFs). The TF activities were estimated using a curated TF-target network80 (IV) and linked to the LRs using a protein–protein interaction network22. B) Mean silhouette scores of clustering heart failure versus healthy myeloid samples using genes extracted from CCC predictions across the distinct steps shown in subpanel A. Silhouette scores were z-transformed using randomly permuted gene sets of the same size as a baseline. The red dashed line signifies a z-score of 1.645, equivalent to a significant one-sided z-test. C) Adjusted Rand Index (ARI) across the distinct steps shown in subpanel A. The central line within each box marks the median, with the box hinges representing the first and third quartiles. The whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Hollow points represent outliers. In the evaluation, Steps 1 and I were respectively represented by prediction averages from CellPhoneDB P-values and SpatialDM’s Global Moran’s R across samples (Methods). D) Histogram of ARI values for Step 6, with one-sided empirical P-values calculated against a random distribution generated using sets of genes of the same size as the predictions in each step. The grey boxes depict the frequency of values drawn at random; the blue dashed line corresponds to the actual value. Source numerical data are available in source data. Source data
Extended Data Fig. 7 LIANA+ generates interpretable and reliable predictions from complex datasets. A) Applying LIANA+ on Myocardial infarction substantially reduced the prediction space to yield an interpretable subset of predictions. (1) inference of the full possible prediction space of ligand-receptor interactions across 110 cell-type pairs, 28 samples, and 3 conditions. (2) Incorporation of spatially associated cell-type pairs (III). (3) Inclusion of information about spatially-colocalised ligand-receptor interactions (LRs), identified using (I) local scores with non-negative matrix factorization (II). (4) Multi-view factor analysis using LRs predicted from single-cell data and informed by the spatial analysis. (5) Targeted differential expression analysis to find significantly deregulated LRs in a specific pair of interacting literature-supported cell types (‡). (6) A causal network search to generate a specific mechanistic hypothesis linking LR interactions to downstream Transcription Factors (TFs). The TF activities were estimated using a curated TF-target network80 (IV) and linked to the LRs using a protein–protein interaction network22. B) Mean silhouette scores of clustering heart failure versus healthy myeloid samples using genes extracted from CCC predictions across the distinct steps shown in subpanel A. Silhouette scores were z-transformed using randomly permuted gene sets of the same size as a baseline. The red dashed line signifies a z-score of 1.645, equivalent to a significant one-sided z-test. C) Adjusted Rand Index (ARI) across the distinct steps shown in subpanel A. The central line within each box marks the median, with the box hinges representing the first and third quartiles. The whiskers extend up to 1.5 times the interquartile range above and below the box hinges. Hollow points represent outliers. In the evaluation, Steps 1 and I were respectively represented by prediction averages from CellPhoneDB P-values and SpatialDM’s Global Moran’s R across samples (Methods). D) Histogram of ARI values for Step 6, with one-sided empirical P-values calculated against a random distribution generated using sets of genes of the same size as the predictions in each step. The grey boxes depict the frequency of values drawn at random; the blue dashed line corresponds to the actual value. Source numerical data are available in source data. Source dataCC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 118 references

Research works printed but not linked: 1

  1. Saarenpää, S. et al. Spatial metatranscriptomics resolves host–bacteria–fungi interactomes. Nat. Biotechnol. https://doi.org/10.1038/s41587-023-01979-2 (2023). · DOI

Cited by

See all 263

Metadata

Authors
Dimitrov, D., Schäfer, P. S. L., Farr, E., Rodriguez-Mier, P., Lobentanzer, S., Badia-i-Mompel, P., Dugourd, A., Tanevski, J., Ramirez Flores, R. O., & Saez-Rodriguez, J.
Affiliations
Heidelberg University (DE)
Venue
Nature Cell Biology
Published
2024-09
Volume
26
Issue
9
Pages
1613-1622
ISSN
  • 1465-7392 (ISSN-L)
  • 1476-4679
OpenAlex
W4402135155
Cited by
263
References
94
Citations by year
2026: 161 · 2025: 91 · 2024: 10
FWCI
35.29
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Cell Biology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
  • Pfizer
  • European Commission (HORIZON2020, 860329, 965193, 951773)
  • Deutsche Forschungsgemeinschaft (SFB1550, SPP 2395, CRC/SFB 1550)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
Keywords
intercellular communication, lianas, single-cell transcriptomics, spatial transcriptomics, multi-omics integration, knowledge base
MeSH terms
Animals, Cell Communication, Humans, Software, Signal Transduction, Computational Biology, Gene Expression Profiling, Single-Cell Analysis, Transcriptome

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: LIANA+ provides an all-in-one framework for cell–cell communication inference. https://app.cochat.ai/papers/10.1038/s41556-024-01469-w
The paper
Dimitrov, D., Schäfer, P. S. L., Farr, E., Rodriguez-Mier, P., Lobentanzer, S., Badia-i-Mompel, P., Dugourd, A., Tanevski, J., Ramirez Flores, R. O., & Saez-Rodriguez, J. (2024). LIANA+ provides an all-in-one framework for cell–cell communication inference. Nature Cell Biology. https://doi.org/10.1038/s41556-024-01469-w

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.