An introduction to representation learning for single-cell data analysis

Gunawan, Indra; Vafaee, Fatemeh; Meijering, Erik; Lock, John G.

doi:10.1016/j.crmeth.2023.100547

Cited by 7 publications

(5 citation statements)

References 105 publications

(108 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, utilizing existing knowledge from annotated datasets could be further exploited with more sophisticated reference-guided cell type annotation algorithms. 19 …”

Section: Discussionmentioning

confidence: 99%

“…Consequently, exploiting annotated datasets using projections to reference datasets (referred hereafter as reference-guided annotation) or other similar approaches that “align” cells to reference datasets is a promising research direction for visualization and automated cell type annotation for large-scale data exploration. 16 , 17 , 18 , 19 , 20 …”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Automated cell type annotation and exploration of single-cell signaling dynamics using mass cytometry

Kleftogiannis,

Gavasso,

Tislevoll

et al. 2024

iScience

View full text Add to dashboard Cite

“…Finally, utilizing existing knowledge from annotated datasets could be further exploited with more sophisticated reference-guided cell type annotation algorithms. 19 …”

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Automated cell type annotation and exploration of single-cell signaling dynamics using mass cytometry

Kleftogiannis,

Gavasso,

Tislevoll

et al. 2024

iScience

View full text Add to dashboard Cite

“…In this way, it would be easier for users with no prior programming knowledge skills to visualise immune landscapes and explore the structure of their datasets facilitating comparisons with the reference. Finally, utilising existing knowledge from annotated datasets following the developed framework is a promising research direction that can be further exploited in the future with more sophisticated reference-guided cell type annotation algorithms (19).…”

Section: Discussionmentioning

confidence: 99%

“…For example, it has been shown that supervised ML algorithms trained with single cells from annotated cell types, achieve high precision (6), whereas semi-supervised approaches (14)(15) gain popularity due to their ability to delineate the structure of the data and identify 'rare' cell types. Consequently, exploiting annotated datasets using projections to reference datasets (referred hereafter as reference-guided annotation) or other similar approaches that 'align' cells to reference datasets is a promising research direction for visualisation and automated cell type annotation for large scale data exploration (16)(17)(18)(19)(20).…”

Section: Introductionmentioning

confidence: 99%

Automated cell type annotation and exploration of single cell signalling dynamics using mass cytometry

Kleftogiannis

Tislevoll

Hellesøy

et al. 2022

Preprint

View full text Add to dashboard Cite

The use of single-cell profiling of phenotypes is suggested to inform about chemoresistance and lack of treatment response in cancer. Mass cytometry by time-of-flight (CyTOF) allows high throughput multiparametric analysis at the single-cell level to perform for in-depth characterisation of heterogeneity in leukemia. However, computational identification of cell populations from CyTOF, and utilisation of single-cell data for biomarker discoveries is challenging. Here, we deployed a machine learning-based framework that enables automatic cell population annotation, and systematic exploration of interactions between signalling proteins in a CyTOF antibody panel. We applied the developed framework to analyse a cohort of 45 leukemia patients. We investigated associations between the cellular composition and clinicopathological and genetic features, and reported salient signalling interactions of Multipotent Progenitor-like leukemia cells that were sufficient to predict short-term survival at time of diagnosis. Our findings confirmed that targeting cell type-specific signalling interactions in leukemia might improve existing patient stratification methods with the potential to inform early about more precise treatment options.

show abstract

“…Many methods have been developed to perform dimensionality reduction, and can be divided into three main groups: statistical methods, manifold methods, and neural networks. [18] This study focused on developing a neural network approach to produce generalizable cell type representations while (1) minimizing batch effects and (2) promoting biological variation. Previously developed neural network algorithms for batch correction have used methods such as autoencoders (AEs) [19, 20], variational autoen-coders (VAEs) [8], generative adversarial networks (GANs) [21], AEs combined with recurrent neural networks (RNNs) [22], contrastive learning [23, 24], and transformers [10].…”

Section: Introductionmentioning

confidence: 99%

Contrastive Learning for Robust Cell Annotation and Representation from Single-Cell Transcriptomics

Andrekson,

Mercado

2024

Preprint

View full text Add to dashboard Cite

Batch effects are a significant concern in single-cell RNA sequencing (scRNA-Seq) data analysis, where variations in the data can be attributed to factors unrelated to cell types. This can make downstream analysis a challenging task. In this study, we present a novel deep learning approach using contrastive learning and a carefully designed loss function for learning an generalizable embedding space from scRNA-Seq data. We call this model CELLULAR: CELLUlar contrastive Learning for Annotation and Representation. When benchmarked against multiple established methods for scRNA-Seq integration, CELLULAR outperforms existing methods in learning a generalizable embedding space on multiple datasets. Cell annotation was also explored as a downstream application for the learned embedding space. When compared against multiple well-established methods, CELLULAR demonstrates competitive performance with top cell classification methods in terms of accuracy, balanced accuracy, and F1 score. CELLULAR is also capable of performing novel cell type detection. These findings aim to quantify themeaningfulnessof the embedding space learned by the model by highlighting the robust performance of our learned cell representations in various applications. The model has been structured into an open-source Python package, specifically designed to simplify and streamline its usage for bioinformaticians and other scientists interested in cell representation learning.

show abstract

An introduction to representation learning for single-cell data analysis

Cited by 7 publications

References 105 publications

Automated cell type annotation and exploration of single-cell signaling dynamics using mass cytometry

Automated cell type annotation and exploration of single-cell signaling dynamics using mass cytometry

Automated cell type annotation and exploration of single cell signalling dynamics using mass cytometry

Contrastive Learning for Robust Cell Annotation and Representation from Single-Cell Transcriptomics

Contact Info

Product

Resources

About