Unsupervised cellular phenotypic hierarchy enables spatial intratumor heterogeneity characterization, recurrence-associated microdomains discovery, and harnesses network biology from hyperplexed in-situ fluorescence images of colorectal carcinoma

Furman, Samantha A.; Stern, Andrew M.; Uttam, Shikhar; Taylor, D. Lansing; Pullara, Filippo; Chennubhotla, S. Chakra

doi:10.1101/2020.10.02.322529

Cited by 1 publication

(4 citation statements)

References 88 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Secondly, cellular neighborhoods (CNs) are defined by Schürch as regions with a characteristic local stoichiometry of cell types. Similar concepts are proposed by other groups, including spatial domains 20 , spatial communities 30 , microdomains 23,31 or tissue domains 32 , tissue niche 33 . There is no uniform definition recognized by the community to define CNs.…”

mentioning

confidence: 78%

“…Python library scikit-learn is used to calculate ARI, and specifically, Python function adjusted_rand_score() is used. Before calculating ARI, we generate the contingency table (23). Given a set 𝑆 of 𝑛 elements, and two clusterings of these elements, namely 𝑋 = {𝑋 , 𝑋 , … , 𝑋 } and 𝑌 = {𝑌 , 𝑌 , … , 𝑌 } , the overlap between 𝑋 and 𝑌 can be summarized in a contingency table 𝑛 where each entry 𝑛 denotes the number of objects in common between 𝑋 and 𝑌 : 𝑛 = |𝑋 ∩ 𝑌 |.…”

Section: Arimentioning

confidence: 99%

“…BayesSpace 22 introduces neighboring spot structure into the prior and update HMRF model with a Bayesian approach to assign clusters. LEAPH 23 uses linear factorization-based soft clustering with spatial regularization to define cell phenotypes. SpiceMix 24 introduces non-negative matrix factorization (NMF) into HMRF framework to jointly model spatial dependency and latent factors of cell identity, leading to novel cell subtypes discovery from mouse visual cortex data.…”

mentioning

confidence: 99%

“…HistoCAT 35 uses statistical permutation test to identify patterns of pairwise cellular colocalization across a tissue and has been widely adopted in research communities. LEAPH 23 introduces pointwise mutual information as pairwise association statistics to discover microdomains. Both HistoCAT and LEAPH can only detect pairwise co-occurrence at a time and rely on sequential statistics analysis instead of a learning-based approach.…”

mentioning

confidence: 99%

See 3 more Smart Citations

Position-informed contrastive learning for spatially resolved omics deciphers hierarchical tissue structure at both cellular and niche levels

Yang

Wang

Liu

et al. 2022

Preprint

View full text Add to dashboard Cite

Spatial omics and multiplexed imaging technologies provide unprecedented spatial context to characterize molecular variation in complex tissues. Developing unsupervised computational pipeline in a discovery mode is of vital importance to interpret spatially resolved data and derive novel biological insights. We develop Stereo, a unified framework leveraging contrastive learning to jointly model molecular and spatial information in a self-supervised manner. Stereo consists of two modules, StereoCell for neighbor-aware cell identity inference and clustering on spatial transcriptomics data, and StereoNiche for biologically-relevant niche identification on spatial proteomics data. StereoCell uses graph attention to aggregate immediate neighbors’ content for each cell while StereoNiche exploits vision Transformer to explicitly relate position coordinates with marker readouts within certain receptive field. We validate StereoCell’s superior utility on both simulated and real transcriptomics datasets of brain tissues acquired by various technologies with different resolutions. StereoCell can operate on all genes, extract layer-specific signature genes at single-cell resolution and identify clear layer-wise structure with state-of-the-art performance. Applied on tumor tissue sections, StereoCell enables detailed characterization of intratumoral heterogeneity and reveals invasive or metastasis behaviors concordant with expert annotations. Evaluated on a multiplexed fluorescence imaging dataset for colorectal cancer (CRC) patients, StereoNiche refines niche assignments, recapitulates classical follicle structure, and validates enriched T cells-macrophages interdigitation. Via self-distillation operation, StereoNiche also enables extracting bounded layout of tertiary lymphoid structure with no supervision, representing classical Crohn’s-like reaction behavior. Local explanations are derived for a tree-based survival model to validate the clinical relevance of identified niches. For mass spectrometry profiled triple-negative breast cancer (TNBC) samples, StereoNiche recognizes differential tumor-immune interactions to separate compartmentalized from mixed pattern. In brief, Stereo is applicable to a variety of spatial technologies, serving as a powerful method to interrogate context-aware biological process and facilitate hierarchical signature discovery in neuroscience and computational pathology.

show abstract

mentioning

confidence: 78%

Section: Arimentioning

confidence: 99%

mentioning

confidence: 99%

mentioning

confidence: 99%

See 2 more Smart Citations