xCell: digitally portraying the tissue cellular heterogeneity landscape

xCell: digitally portraying the tissue cellular heterogeneity landscape Dvir Aran Zicheng Hu Atul J. Butte

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

xCell: digitally portraying the tissue cellular heterogeneity landscape 2017 Genome biology 10.1186/s13059-017-1349-1 aran2017-xcell-digitally-portraying-the-tis aran2017-xcell-digitally-portraying-the-tis.txt, the project's staged copy 6bf7a0f2984e3e44e3d6725310cdcd03bada5313 cc-by biology

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them inferring which cell types are present in a tissue from its bulk gene expression - and how far such an inference can be trusted Quotation withheld because this record is not cleared for redistribution. whether enrichment scores can be made comparable across cell types and freed of the correlation between closely related ones, which is what signature methods could not previously do Quotation withheld because this record is not cleared for redistribution. xCell - 489 signatures for 64 cell types chosen on held-out data, a power-law calibration onto a common scale, a spillover compensation borrowed from flow cytometry, and a null distribution for absence Quotation withheld because this record is not cleared for redistribution. method Quotation withheld because this record is not cleared for redistribution. a measurement that returns a non-zero score for cell types that are not in the sample, and scores that look like proportions without being them Quotation withheld because this record is not cleared for redistribution. system-or-tool-building system-or-tool-building Quotation withheld because this record is not cleared for redistribution. closely related cell types share marker genes, so a signature for one is partly satisfied by the other - formally the same problem as fluorophores whose emission spectra overlap, and correctable by the same device, a compensation matrix estimated from mixtures where the true answer is known spectral overlap as a statistical analogy Quotation withheld because this record is not cleared for redistribution. before compensation each cell type's score is computed alone, so one wrong inference cannot contaminate the rest; the correction couples them, which is exactly the property that makes deconvolution fragile - the paper accepts a limited amount of it and says how limited compensation trading independence for separability Quotation withheld because this record is not cleared for redistribution. a signature returns a score for any input, so a cell type absent from the mixture still receives one; only a null distribution built from mixtures that exclude it can say whether the score means presence unbounded scores for absent cell types Quotation withheld because this record is not cleared for redistribution. signatures are learned from cells isolated and profiled alone, and applied to cells sitting in tissue; if marker expression differs between those two states the method is calibrated on a population it never meets purified profiles differing from the same cells in a mixture Quotation withheld because this record is not cleared for redistribution. the outputs are enrichment scores and cannot be interpreted as proportions, although they are calibrated to resemble them Quotation withheld because this record is not cleared for redistribution. every method, this one included, performs significantly better on simulated mixtures than on real ones, and the real-data correlations are called far from perfect Quotation withheld because this record is not cleared for redistribution. 56.4% of scores for cell types absent from a test mixture were non-negligible; the significance test brings that to 28.8% and costs 13.5 points of the true positives Quotation withheld because this record is not cleared for redistribution. the spillover correction removes the independence between cell type inferences that was the signature approach's main protection - limited, and the authors say so Quotation withheld because this record is not cleared for redistribution. on CD8+ T cells the method is beaten by using CD8A expression alone, with the caveat that the single gene may not be reliable in cancer Quotation withheld because this record is not cleared for redistribution. for five cell types the null distribution degenerated to zero and was replaced by a hand-set constant of 0.001 Quotation withheld because this record is not cleared for redistribution. spillover compensation, carried from flow cytometry - where fluorophore emission spectra overlap - into gene signature scoring, where marker genes overlap Quotation withheld because this record is not cleared for redistribution. null-distribution fitting for a presence test, carried from statistical hypothesis testing to ask whether a cell type is in a mixture at all Quotation withheld because this record is not cleared for redistribution. signatures selected only if they hold up in a data source they were NOT trained on eliminates: overfitting to the source a signature was learned from as the explanation of its performance - the reliance on reference data that the paper names as the first of the field's seven problems · outcome: held Quotation withheld because this record is not cleared for redistribution. simulated mixtures built from an independent data source never used in development (GSE60424), scored before and after spillover compensation eliminates: the development data as the source of the improvement - a correction tuned on the training sources could otherwise be measuring itself · outcome: held Quotation withheld because this record is not cleared for redistribution. comparison against 53 previously published signatures and against CIBERSORT, a deconvolution method, on the same mixtures eliminates: the possibility that any signature set would score as well - the comparison is to named alternatives on identical inputs rather than to nothing · outcome: held Quotation withheld because this record is not cleared for redistribution. validation against cytometry immunophenotyping in 165 individuals across two independent studies, rather than against simulations alone eliminates: in-silico mixing as the reason the method works - arithmetic admixtures cannot test whether marker expression survives being in real tissue · outcome: partial Quotation withheld because this record is not cleared for redistribution. a null distribution per cell type, learned from random mixtures that EXCLUDE that cell type and its relatives, to test whether a score means presence at all eliminates: a non-zero score as evidence of presence - without this the method reports something for every cell type in every sample · outcome: partial Quotation withheld because this record is not cleared for redistribution. the significance threshold set at 20% and the trade-off stated, rather than chosen silently eliminates: a threshold tuned to flatter the result - the cost in true-positive power is reported alongside the gain in false-positive suppression · outcome: held Quotation withheld because this record is not cleared for redistribution. enrichment scores enrichment scores Quotation withheld because this record is not cleared for redistribution. proportions proportions Quotation withheld because this record is not cleared for redistribution. spillover effect spillover effect Quotation withheld because this record is not cleared for redistribution. deconvolution deconvolution Quotation withheld because this record is not cleared for redistribution. gene signatures gene signatures Quotation withheld because this record is not cleared for redistribution. simulated mixtures simulated mixtures Quotation withheld because this record is not cleared for redistribution. real mixtures real mixtures Quotation withheld because this record is not cleared for redistribution. tumor microenvironment tumor microenvironment Quotation withheld because this record is not cleared for redistribution. CD8A CD8A Quotation withheld because this record is not cleared for redistribution. ssGSEA ssGSEA Quotation withheld because this record is not cleared for redistribution. xCell's performance - IN SIMULATED MIXTURES: outperforms every compared method in the large majority of cell types, across data sources and platforms Quotation withheld because this record is not cleared for redistribution. xCell's performance - IN REAL MIXTURES: significantly better than the alternatives and still, in the authors' words, far from perfect - and beaten outright on CD8+ T cells by using one gene Quotation withheld because this record is not cleared for redistribution. an xCell score - AS COMPUTED: a single-sample gene-set enrichment score, interpretable only against other samples Quotation withheld because this record is not cleared for redistribution. an xCell score - AS CALIBRATED: deliberately transformed to resemble a proportion so cell types can be compared, and explicitly not usable as a proportion Quotation withheld because this record is not cleared for redistribution. critiques → in-silico cell-type dissection methods generally, on seven named grounds including reliance on the reference source, narrow cell-type coverage, and validation on few cell types in few samples Quotation withheld because this record is not cleared for redistribution. qualifies → CIBERSORT and other deconvolution approaches, whose strict inter-cell-type dependencies make them fragile on tumour expression profiles and tied to their reference matrix Quotation withheld because this record is not cleared for redistribution. extends → gene set enrichment analysis, by calibrating its scores onto a common scale and compensating the correlation between related signatures Quotation withheld because this record is not cleared for redistribution. qualifies → its own results and every comparable method's - digital dissection findings must be validated by other technologies before they are believed Quotation withheld because this record is not cleared for redistribution. enrichment scores —distinct-from→ proportions enrichment scores proportions differs in: whether the value can be read as a share of the mixture · the relation itself is separately evidenced · strength of the relation: 0.95 · differs in (stated by the tagger, no quotation): whether the value can be read as a share of the mixture · the relation stated: the score is calibrated to resemble a fraction and cannot be used as one Quotation withheld because this record is not cleared for redistribution. simulated mixtures —contrasted-with→ real mixtures simulated mixtures real mixtures differs in: whether the cells were mixed by arithmetic or by biology · the relation itself is separately evidenced · strength of the relation: 0.95 · differs in (stated by the tagger, no quotation): whether the cells were mixed by arithmetic or by biology · the relation stated: every method scores better on in-silico admixtures than on tissue Quotation withheld because this record is not cleared for redistribution. every method, including this one, performs significantly better on simulated mixtures than on real ones every method, including this one, performs significantly better on simulated mixtures than on real ones Quotation withheld because this record is not cleared for redistribution. the scores are calibrated to resemble proportions and cannot be used as proportions the scores are calibrated to resemble proportions and cannot be used as proportions Quotation withheld because this record is not cleared for redistribution. 56.4% of scores for cell types absent from a mixture were non-negligible before the significance test, and 28.8% after 56.4% of scores for cell types absent from a mixture were non-negligible before the significance test, and 28.8% after Quotation withheld because this record is not cleared for redistribution. Spillover correction removes the strict independence between cell-type inferences, though compensation is limited and most inferences stay independent. Spillover correction removes the strict independence between cell-type inferences, though compensation is limited and most inferences stay independent. Quotation withheld because this record is not cleared for redistribution. on CD8+ T cells a single gene outperforms the whole method, with the caveat that the gene may be unreliable in cancer on CD8+ T cells a single gene outperforms the whole method, with the caveat that the gene may be unreliable in cancer Quotation withheld because this record is not cleared for redistribution. a cell type's score showing enrichment because a closely related cell type is present - the same shape as spectral overlap between fluorophores in flow cytometry spillover effect Quotation withheld because this record is not cleared for redistribution. the inferences this method produces - strictly scores, calibrated to resemble proportions and explicitly not interpretable as them enrichment scores Quotation withheld because this record is not cleared for redistribution. scores for cell types absent from the mixture that were nonetheless non-negligible 56.4% Quotation withheld because this record is not cleared for redistribution. the share of genes xCell requires that were actually present in one validation study 54.5% Quotation withheld because this record is not cleared for redistribution. the significance threshold, chosen as a stated trade-off 20% Quotation withheld because this record is not cleared for redistribution. ssGSEA ssGSEA Quotation withheld because this record is not cleared for redistribution. p value p value Quotation withheld because this record is not cleared for redistribution. enrichment scores enrichment scores Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1186/s13059-017-1349-1.