Generalizable and Scalable Visualization of Single-Cell Data Using Neural Networks

Cho, Hyunghoon; Berger, Bonnie; Peng, Jian

doi:10.1016/j.cels.2018.05.017

Cited by 50 publications

(39 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One potential advantage of this approach is that the 'most appropriate' perplexity does not need to grow with the sample size, as long as the mini-batch size remains constant. Parametric t-SNE has been recently applied to transcriptomic data under the names net-SNE (Cho et al, 2018) and scvis (Ding et al, 2018). The latter method combined parametric t-SNE with a variational autoencoder, and was claimed to yield more interpretable visualisations than standard t-SNE due to better preserving the global structure.…”

Section: Comparison To Related Workmentioning

confidence: 99%

The art of using t-SNE for single-cell transcriptomics

Kobak

Berens

2018

Preprint

137

View full text Add to dashboard Cite

Single-cell transcriptomics yields ever growing data sets containing RNA expression levels for thousands of genes from up to millions of cells. Common data analysis pipelines include a dimensionality reduction step for visualising the data in two dimensions, most frequently performed using t-distributed stochastic neighbour embedding (t-SNE). It excels at revealing local structure in high-dimensional data, but naive applications often suffer from severe shortcomings, e.g. the global structure of the data is not represented accurately. Here we describe how to circumvent such pitfalls, and develop a protocol for creating more faithful t-SNE visualisations. It includes PCA initialisation, a high learning rate, and multi-scale similarity kernels; for very large data sets, we additionally use exaggeration and downsampling-based initialisation. We use published single-cell RNA-seq data sets to demonstrate that this protocol yields superior results compared to the naive application of t-SNE.

show abstract

Section: Comparison To Related Workmentioning

confidence: 99%

The art of using t-SNE for single-cell transcriptomics

Kobak

Berens

2018

Preprint

137

View full text Add to dashboard Cite

show abstract

“…Neural networks provide a popular framework for machine learning algorithms which can be used to interpret complex datasets. As a result, neural networks have been widely used in many fields, including for the analysis of scRNA-seq data [2-5] . Since the output data from scRNA-seq is feature-enriched and well-structured, it is well suited as an input for neural networks.…”

Section: Introductionmentioning

confidence: 99%

Automated identification of Cell Types in Single Cell RNA Sequencing

Pellegrini

2019

Preprint

View full text Add to dashboard Cite

Cell type identification is one of the major goals in single cell RNA sequencing (scRNA-seq). Current methods for assigning cell types typically involve the use of unsupervised clustering, the identification of signature genes in each cluster, followed by a manual lookup of these genes in the literature and databases to assign cell types. However, there are several limitations associated with these approaches, such as unwanted sources of variation that influence clustering and a lack of canonical markers for certain cell types. Here, we present ACTINN (Automated Cell Type Identification using Neural Networks), which employs a neural network with 3 hidden layers, trains on datasets with predefined cell types, and predicts cell types for other datasets based on the trained parameters. We trained the neural network on a mouse cell type atlas (Tabula Muris Atlas) and a human immune cell dataset, and used it to predict cell types for mouse leukocytes, human PBMCs and human T cell sub types. The results showed that our neural network is fast and accurate, and should therefore be a useful tool to complement existing scRNA-seq pipelines.Author SummarySingle cell RNA sequencing (scRNA-seq) provides high resolution profiling of the transcriptomes of individual cells, which inevitably results in high volumes of data that require complex data processing pipelines. Usually, one of the first steps in the analysis of scRNA-seq is to assign individual cells to known cell types. To accomplish this, traditional methods first group the cells into different clusters, then find marker genes, and finally use these to manually assign cell types for each cluster. Thus these methods require prior knowledge of cell type canonical markers, and some level of subjectivity to make the cell type assignments. As a result, the process is often laborious and requires domain specific expertise, which is a barrier for inexperienced users. By contrast, our neural network ACTINN automatically learns the features for each predefined cell type and uses these features to predict cell types for individual cells. This approach is computationally efficient and requires no domain expertise of the tissues being studied. We believe ACTINN allows users to rapidly identify cell types in their datasets, thus rendering the analysis of their scRNA-seq datasets more efficient.

show abstract

“…Several attempts to successfully apply t-SNE-like methods to massive datasets have been recently reported including aforenoted HSNE 7,33,34 , LargeVis 10 and net-SNE 35 . However, these improved methods, when applied to large datasets, often require/benefit from considerable computational resources; for instance, the LargeVis study was performed on a 512Gb RAM, 32 core station.…”

Section: Discussionmentioning

confidence: 99%

Automated optimized parameters for t-distributed stochastic neighbor embedding improve visualization and allow analysis of large datasets

Belkina

Ciccolella

Anno

et al. 2018

Preprint

View full text Add to dashboard Cite

10 11 12 Accurate and comprehensive extraction of information from high-dimensional single cell 13 datasets necessitates faithful visualizations to assess biological populations. A state-of-the-art 14 algorithm for non-linear dimension reduction, t-SNE, requires multiple heuristics and fails to 15 produce clear representations of datasets when millions of cells are projected. We developed 16 opt-SNE, an automated toolkit for t-SNE parameter selection that utilizes Kullback-Liebler 17 divergence evaluation in real time to tailor the early exaggeration and overall number of 18 gradient descent iterations in a dataset-specific manner. The precise calibration of early 19 exaggeration together with opt-SNE adjustment of gradient descent learning rate dramatically 20 improves computation time and enables high-quality visualization of large cytometry and 21 transcriptomics datasets, overcoming limitations of analysis tools with hard-coded parameters 22

show abstract

Generalizable and Scalable Visualization of Single-Cell Data Using Neural Networks

Cited by 50 publications

References 42 publications

The art of using t-SNE for single-cell transcriptomics

The art of using t-SNE for single-cell transcriptomics

Automated identification of Cell Types in Single Cell RNA Sequencing

Automated optimized parameters for t-distributed stochastic neighbor embedding improve visualization and allow analysis of large datasets

Contact Info

Product

Resources

About