CoChat

Style transfer with variational autoencoders is a promising approach to RNA-Seq data harmonization and analysis

Open access
Russkikh, Nikolai, Antonets, Denis, Shtokalo, Dmitry, Makarov, Alexander, Vyatkin, Yuri, Zakharov, Alexey, Terentyev, Evgeny · 2020 · Bioinformatics

Published version of a preprint

The DOI matches its Crossref record; 37 of the 37 references it cites were found in a registry or an index.

Abstract

Motivation: The transcriptomic data are being frequently used in the research of biomarker genes of different diseases and biological states. The most common tasks there are the data harmonization and treatment outcome prediction. Both of them can be addressed via the style transfer approach. Either technical factors or any biological details about the samples which we would like to control (gender, biological state, treatment, etc.) can be used as style components. Results: The proposed style transfer solution is based on Conditional Variational Autoencoders, Y-Autoencoders and adversarial feature decomposition. To quantitatively measure the quality of the style transfer, neural network classifiers which predict the style and semantics after training on real expression were used. Comparison with several existing style-transfer based approaches shows that proposed model has the highest style prediction accuracy on all considered datasets while having comparable or the best semantics prediction accuracy. Availability and implementation: https://github.com/NRshka/stvae-source. Supplementary information: Supplementary data are available at Bioinformatics online.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1093/bioinformatics/btaa624

Cites

37 of 37 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1. The autoencoder architecture scheme
Fig. 1. The autoencoder architecture schemeCC BY-NC · PubMed Central
Fig. 2. Visualization of original samples with tSNE. Raw expression values were used, samples were colored according to cell types (A) and physiological state (B). tSNE perplexity was set to 30
Fig. 2. Visualization of original samples with tSNE. Raw expression values were used, samples were colored according to cell types (A) and physiological state (B). tSNE perplexity was set to 30CC BY-NC · PubMed Central
Fig. 3. Visualization of reconstructed samples with tSNE. Gene expression values reconstructed with VAE model were used, samples were colored according to cell types (A) and physiological state (B). tSNE perplexity was set to 30
Fig. 3. Visualization of reconstructed samples with tSNE. Gene expression values reconstructed with VAE model were used, samples were colored according to cell types (A) and physiological state (B). tSNE perplexity was set to 30CC BY-NC · PubMed Central
Fig. 4. Visualization of the samples with tSNE using the learned latent representation. The latent variables of the testing samples were obtained with pretrained encoder. The samples were colored according to cell types (A) and physiological state (B). tSNE perplexity was set to 30
Fig. 4. Visualization of the samples with tSNE using the learned latent representation. The latent variables of the testing samples were obtained with pretrained encoder. The samples were colored according to cell types (A) and physiological state (B). tSNE perplexity was set to 30CC BY-NC · PubMed Central
Fig. 5. MA-plots comparing the gene expression in stromal cells from murine mammary glands in original and transformed samples. The comparison of original samples is shown with blue; the comparison between the original virgin state and the virgin state produced from pregnancy with style transfer is shown with orange color
Fig. 5. MA-plots comparing the gene expression in stromal cells from murine mammary glands in original and transformed samples. The comparison of original samples is shown with blue; the comparison between the original virgin state and the virgin state produced from pregnancy with style transfer is shown with orange colorCC BY-NC · PubMed Central
Fig. 6. Variation in gene expression related to mammary gland development (GO:0030879) in stromal and luminal cells (A) and in all cells (B)
Fig. 6. Variation in gene expression related to mammary gland development (GO:0030879) in stromal and luminal cells (A) and in all cells (B)CC BY-NC · PubMed Central
Fig. 7. Variation in gene expression related to positive regulation of apoptotic process (GO:0043065) in stromal, luminal and alveolar cells (A) and in all cells (B)
Fig. 7. Variation in gene expression related to positive regulation of apoptotic process (GO:0043065) in stromal, luminal and alveolar cells (A) and in all cells (B)CC BY-NC · PubMed Central
Fig. 8. Results of GO-enrichment analysis of top 100 differentially expressed genes between different conditions. All the combinations of original and transfer conditions, models were tested—negative expression values were set to zero, then top 100 differentially expressed genes were selected using Welch’s test and analyzed for GO-terms enrichment. (A) GO-terms enrichment of genes differentially expressed in O(C) and O(S) CD4+ T-cells—from the original data; (B) GO-terms enrichment of genes differentially expressed in original unstimulated CD4+ T-cells—O(C)—and in T(S->C)—the interferon-treated CD4+ T-cells that were turned into unstimulated condition with our style-transfer procedure. (C) A ‘birds-eye’ view of the table with results of all-to-all comparison in terms of differentially expressed genes overlap and GO-terms enrichment analysis. The yellow color indicates comparisons of samples whose conditions were different either originally or after the style transfer. The four colored columns correspond to (i) the Jaccard index between top 100 differentially expressed genes found in current samples combination and in O(S) versus O(C) (chosen as etalon); (ii) a number of significantly enriched GO categories; (iii) the Jaccard index between top 10 GO categories (sorted according to enrichment significance) against the top 10 GOs found in O(S) versus O(C) DEG comparison and (iv) the Jaccard index between top 30 GO categories and the top 30 GOs in O(S) versus O(C) comparison. The color varies from red to blue corresponding to value changes from low to high. The original table can be found in Supplementary Table ST6
Fig. 8. Results of GO-enrichment analysis of top 100 differentially expressed genes between different conditions. All the combinations of original and transfer conditions, models were tested—negative expression values were set to zero, then top 100 differentially expressed genes were selected using Welch’s test and analyzed for GO-terms enrichment. (A) GO-terms enrichment of genes differentially expressed in O(C) and O(S) CD4+ T-cells—from the original data; (B) GO-terms enrichment of genes differentially expressed in original unstimulated CD4+ T-cells—O(C)—and in T(S->C)—the interferon-treated CD4+ T-cells that were turned into unstimulated condition with our style-transfer procedure. (C) A ‘birds-eye’ view of the table with results of all-to-all comparison in terms of differentially expressed genes overlap and GO-terms enrichment analysis. The yellow color indicates comparisons of samples whose conditions were different either originally or after the style transfer. The four colored columns correspond to (i) the Jaccard index between top 100 differentially expressed genes found in current samples combination and in O(S) versus O(C) (chosen as etalon); (ii) a number of significantly enriched GO categories; (iii) the Jaccard index between top 10 GO categories (sorted according to enrichment significance) against the top 10 GOs found in O(S) versus O(C) DEG comparison and (iv) the Jaccard index between top 30 GO categories and the top 30 GOs in O(S) versus O(C) comparison. The color varies from red to blue corresponding to value changes from low to high. The original table can be found in Supplementary Table ST6CC BY-NC · PubMed Central
Fig. 9. The original expression (decimal logarithm values) and the expression reconstructed from autoencoder after style-transfer in genes, associated with type I interferon signaling pathway, in CD4+ T-cells. (A) The reconstructed gene expression in CD4+ T-cells after transformation from interferon-stimulated state into unstimulated (control) state (S->C) and after the transformation from control to interferon-stimulated state (C->S); (B) the comparison of original gene expression in unstimulated CD4+ T-cells—Original(C)—against the expression obtained with transformation from interferon-stimulated into control state; (C) the comparison of original gene expression in unstimulated—Original(C)—against stimulated CD4+ T-cells—Original(S); (D) the original gene expression in unstimulated CD4+ T-cells—Original(C)—against the style transfer C->S obtained with the model trained without the original stimulated CD4+ T-cells gene expression data; (E) the transformations from control to stimulated state were done using a model pretrained on the whole training GSE96583 dataset (tf stimulated) or with a model trained on a subset of the training samples without stimulated CD4+ T-cells [tf stimulated wo CD4T(S)]; (F) the S->C and S->C transformations from a model trained without seeing the stimulated CD4+ T-cells expression data. The validation subset of GSE96583 data was used
Fig. 9. The original expression (decimal logarithm values) and the expression reconstructed from autoencoder after style-transfer in genes, associated with type I interferon signaling pathway, in CD4+ T-cells. (A) The reconstructed gene expression in CD4+ T-cells after transformation from interferon-stimulated state into unstimulated (control) state (S->C) and after the transformation from control to interferon-stimulated state (C->S); (B) the comparison of original gene expression in unstimulated CD4+ T-cells—Original(C)—against the expression obtained with transformation from interferon-stimulated into control state; (C) the comparison of original gene expression in unstimulated—Original(C)—against stimulated CD4+ T-cells—Original(S); (D) the original gene expression in unstimulated CD4+ T-cells—Original(C)—against the style transfer C->S obtained with the model trained without the original stimulated CD4+ T-cells gene expression data; (E) the transformations from control to stimulated state were done using a model pretrained on the whole training GSE96583 dataset (tf stimulated) or with a model trained on a subset of the training samples without stimulated CD4+ T-cells [tf stimulated wo CD4T(S)]; (F) the S->C and S->C transformations from a model trained without seeing the stimulated CD4+ T-cells expression data. The validation subset of GSE96583 data was usedCC BY-NC · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 1

  1. Sohn, 2015, Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 2 (NIPS’15), 3483

Cited by

See all 26

Metadata

Authors
Russkikh, N., Antonets, D., Shtokalo, D., Makarov, A., Vyatkin, Y., Zakharov, A., & Terentyev, E.
Affiliations
  • AcademGene LLC , Novosibirsk 630090, Russia
  • Siberian Branch of the Russian Academy of Sciences (RU)
  • Cancer Research Foundation , Moscow 109316, Russia
Venue
Bioinformatics
Published
2020-07-16
Volume
36
Issue
20
Pages
5076-5085
ISSN
  • 1367-4803 (ISSN-L)
  • 1367-4811
OpenAlex
W3042427971
Cited by
21
References
35
Citations by year
2026: 3 · 2025: 6 · 2024: 4 · 2023: 4 · 2022: 3 · 2021: 1
FWCI
0.65
Type
journal article
Language
en
Open access
Locations
  • Bioinformatics (publishedVersion, cc-by-nc)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
Funders
Siberian Branch, Russian Academy of Sciences
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene expression and cancer classification
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
Keywords
style transfer, variational autoencoder, transcriptomics, conditional variational autoencoder, semantic prediction, treatment outcome prediction, biomarker genes
MeSH terms
Exome Sequencing, RNA-Seq, Semantics, Software, Neural Networks, Computer

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.