Probabilistic outlier identification for RNA sequencing generalized linear models

Mangiola, Stefano; Thomas, Evan A.; Modrák, Martin; Vehtari, Aki; Papenfuss, Anthony T.

doi:10.1093/nargab/lqab005

Cited by 9 publications

(9 citation statements)

References 34 publications

(37 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We show that the data property which causes LSCO and RidgeCO to produce megahubs is found in biological data. This is further supported by previous work studying outlier detection and handling in both microarray ( Kadota et al , 2003 ; Shieh and Hung, 2009 ; Yang et al , 2009 ) and sequencing data ( Love et al , 2014 ; Mangiola et al , 2021 ). For GRNI, an example of a cause for such an outlier is the inclusion of a gene that is not part of the studied system.…”

Section: Discussionsupporting

confidence: 80%

Fast and accurate gene regulatory network inference by normalized least squares regression

et al. 2022

View full text Add to dashboard Cite

Motivation Inferring an accurate gene regulatory network (GRN) has long been a key goal in the field of systems biology. In order to do this, it is important to find a suitable balance between the maximum number of true positive and the minimum number of false positive interactions. Another key feature is that the inference method can handle the large size of modern experimental data, meaning the method needs to be both fast and accurate. The LSCO (Least Squares Cut-Off) method can fulfill both these criteria, however as it is based on least squares it is vulnerable to known issues of amplifying extreme values, small or large. In GRN this manifests itself with genes that are erroneously hyper-connected to a large fraction of all genes due to extremely low value fold changes. Results We developed a GRN inference method called LSCON (Least Squares Cut-Off with Normalization) that tackles this problem. LSCON extends the LSCO algorithm by regularization to avoid hyper-connected genes and thereby reduce false positives. The regularization employed is based on normalization, which removes effects of extreme values on the fit. We benchmarked LSCON and compared it to Genie3, LASSO, LSCO, and Ridge regression, in terms of accuracy, speed, and tendency to predict hyper-connected genes. The results show that LSCON achieves better or equal accuracy compared to LASSO, the best existing method, especially for data with extreme values. Thanks to the speed of least squares regression, LSCON does this an order of magnitude faster than LASSO. Availability Data Data: https://bitbucket.org/sonnhammergrni/lscon; Code: https://bitbucket.org/sonnhammergrni/genespider Supplementary information Supplementary data are available at Bioinformatics online.

show abstract

Section: Discussionsupporting

confidence: 80%

Fast and accurate gene regulatory network inference by normalized least squares regression

et al. 2022

View full text Add to dashboard Cite

show abstract

“…The metastatic condition (non-metastatic, low or high-burden) was chosen as the only covariate. We detected outlier sample-gene observations using ppcseq [86]. The significant genes that included outliers were filtered out.…”

Section: Methodsmentioning

confidence: 99%

The circulating immune cell landscape stratifies metastatic burden in breast cancer patients

Mangiola,

Brown,

Berthelet

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Advanced breast cancers show varying degrees of metastasis; however, reliable biomarkers of metastatic disease progression remain unknown. In circulation, immune cells are the first line of defence against tumour cells. Herein, using >109,591 peripheral blood mononuclear cells from healthy individuals and breast cancer patients, we tested whether molecular traits of the circulating immune cells, probed with single-cell transcriptomics, can be used to segregate metastatic profiles. Our analyses revealed significant compositional and transcriptional differences in PBMCs of patients with restricted or high metastatic burden versus healthy subjects. The abundance of T cell and monocyte subtypes segregated cancer patients from healthy individuals, while memory and unconventional T cells were enriched in low metastatic burden disease. The cell communication axes were also found to be tightly associated with the extent of metastatic burden. Additionally, we identified a PBMC-derived metastatic gene signature capable of discerning metastatic condition from a healthy state. Our study provides unique molecular insights into the peripheral immune system operating in metastatic breast cancer, revealing potential new biomarkers of the extent of the metastatic state. Tracking such immune traits associated with metastatic spread could complement existing diagnostic tools.

show abstract

“…This fit will likely be not biased by outliers and can produce reliable posterior probability distribution to base an accurate outlier identification. The posterior predictive distribution is then produced adjusting for observation censoring (42). This adjustment is necessary because eliminating data at the distribution’s tails leads to downwards biases for the estimated variance.…”

Section: Methodsmentioning

confidence: 99%

“…A robust iterative strategy for outlier identification was developed for negative-binomial data from bulk RNA sequencing (42). Such a strategy is necessary because a fit that includes outliers makes the model biased by definition and produces skewed estimates.…”

Section: Methodsmentioning

confidence: 99%

Robust differential composition and variability analysis for multisample cell omics

Mangiola

Schulze

Trussart

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Cell omics such as single-cell genomics, proteomics and microbiomics allow the characterisation of tissue and microbial community composition, which can be compared between conditions to identify biological drivers. This strategy has been critical to unveiling markers of disease progression such as cancer and pathogen infection. For cell omic data, no method for differential variability analysis exists, and methods for differential composition analysis only take a few fundamental data properties into account. Here we introduce sccomp, a generalised method for differential composition and variability analyses able to jointly model data count distribution, compositionality, group-specific variability and proportion mean-variability association, with awareness against outliers. Sccomp is an extensive analysis framework that allows realistic data simulation and cross-study knowledge transfer. Here, we demonstrate that mean-variability association is ubiquitous across technologies showing the inadequacy of the very popular Dirichlet-multinomial modelling and provide mandatory principles for differential variability analysis. We show that sccomp accurately fits experimental data, with a 50% incremental improvement over state-of-the-art algorithms. Using sccomp, we identified novel differential constraints and composition in the microenvironment of primary breast cancer.Significance statementDetermining the composition of cell populations is made possible by technologies like single-cell transcriptomics, CyTOF and microbiome sequencing. Such analyses are now widespread across fields (~800 publications/month, Scopus). However, existing methods for differential abundance do not model all data features, and cell-type/taxa specific differential variability is not yet possible. Increase in the variability of tissue composition and microbial communities is a well-known indicator of loss of homeostasis and disease. A suitable statistical method would enable new types of analyses to identify component-specific loss of homeostasis for the first time. This and other innovations are now possible through our discovery of the mean-variability association for compositional data. Based on this fundamental observation, we have developed a new statistical model, sccomp, that enables differential variability analysis for composition data, improved differential abundance analyses, with cross-sample information borrowing, outlier identification and exclusion, realistic data simulation, based on experimental datasets, cross-study knowledge transfer.

show abstract

Probabilistic outlier identification for RNA sequencing generalized linear models

Cited by 9 publications

References 34 publications

Fast and accurate gene regulatory network inference by normalized least squares regression

Fast and accurate gene regulatory network inference by normalized least squares regression

The circulating immune cell landscape stratifies metastatic burden in breast cancer patients

Robust differential composition and variability analysis for multisample cell omics

Contact Info

Product

Resources

About