DANCE: a deep learning library and benchmark platform for single-cell analysis DOI Creative Commons
Jiayuan Ding, Renming Liu, Hongzhi Wen

et al.

Genome biology, Journal Year: 2024, Volume and Issue: 25(1)

Published: March 19, 2024

DANCE is the first standard, generic, and extensible benchmark platform for accessing evaluating computational methods across spectrum of datasets numerous single-cell analysis tasks. Currently, supports 3 modules 8 popular tasks with 32 state-of-art on 21 datasets. People can easily reproduce results supported algorithms major via minimal efforts, such as using only one command line. In addition, provides an ecosystem deep learning architectures tools researchers to facilitate their own model development. open-source Python package that welcomes all kinds contributions.

Language: Английский

Joint variational autoencoders for multimodal imputation and embedding DOI
Noah Cohen Kalafut, Xiang Huang, Daifeng Wang

et al.

Nature Machine Intelligence, Journal Year: 2023, Volume and Issue: 5(6), P. 631 - 642

Published: May 29, 2023

Language: Английский

Citations

26
siVAE: interpretable deep generative models for single-cell transcriptomes DOI Creative Commons
Yongin Choi, Ruoxin Li, Gerald Quon

et al.

Genome biology, Journal Year: 2023, Volume and Issue: 24(1)

Published: Feb. 20, 2023

Abstract Neural networks such as variational autoencoders (VAE) perform dimensionality reduction for the visualization and analysis of genomic data, but are limited in their interpretability: it is unknown which data features represented by each embedding dimension. We present siVAE, a VAE that interpretable design, thereby enhancing downstream tasks. Through interpretation, siVAE also identifies gene modules hubs without explicit network inference. use to identify whose connectivity associated with diverse phenotypes iPSC neuronal differentiation efficiency dementia, showcasing wide applicability generative models analysis.

Language: Английский

Citations

25
scButterfly: a versatile single-cell cross-modality translation method via dual-aligned variational autoencoders DOI Creative Commons

Yichuan Cao,

Xiamiao Zhao,

Songming Tang

et al.

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: April 6, 2024

Abstract Recent advancements for simultaneously profiling multi-omics modalities within individual cells have enabled the interrogation of cellular heterogeneity and molecular hierarchy. However, technical limitations lead to highly noisy multi-modal data substantial costs. Although computational methods been proposed translate single-cell across modalities, broad applications still remain impeded by formidable challenges. Here, we propose scButterfly, a versatile cross-modality translation method based on dual-aligned variational autoencoders augmentation schemes. With comprehensive experiments multiple datasets, provide compelling evidence scButterfly’s superiority over baseline in preserving while translating datasets various contexts revealing cell type-specific biological insights. Besides, demonstrate extensive scButterfly integrative analysis single-modality data, enhancement poor-quality multi-omics, automatic type annotation scATAC-seq data. Moreover, can be generalized unpaired training, perturbation-response analysis, consecutive translation.

Language: Английский

Citations

12
scmFormer Integrates Large‐Scale Single‐Cell Proteomics and Transcriptomics Data by Multi‐Task Transformer DOI Creative Commons
Jing Xu,

De‐Shuang Huang,

Xiujun Zhang

et al.

Advanced Science, Journal Year: 2024, Volume and Issue: 11(19)

Published: March 14, 2024

Abstract Transformer‐based models have revolutionized single cell RNA‐seq (scRNA‐seq) data analysis. However, their applicability is challenged by the complexity and scale of single‐cell multi‐omics data. Here a novel multi‐modal/multi‐task transformer (scmFormer) proposed to fill up existing blank integrating proteomics with other omics Through systematic benchmarking, it demonstrated that scmFormer excels in large‐scale multimodal heterogeneous multi‐batch paired data, while preserving shared information across batchs distinct biological information. achieves 54.5% higher average F1 score compared second method transferring cell‐type labels from transcriptomics Using COVID‐19 datasets, presented successfully integrates over 1.48 million cells on personal computer. Moreover, also proved performs better than methods generating unmeasured modality well‐suited for spatial multi‐omic Thus, powerful comprehensive tool analyzing

Language: Английский

Citations

10
DANCE: a deep learning library and benchmark platform for single-cell analysis DOI Creative Commons
Jiayuan Ding, Renming Liu, Hongzhi Wen

et al.

Genome biology, Journal Year: 2024, Volume and Issue: 25(1)

Published: March 19, 2024

DANCE is the first standard, generic, and extensible benchmark platform for accessing evaluating computational methods across spectrum of datasets numerous single-cell analysis tasks. Currently, supports 3 modules 8 popular tasks with 32 state-of-art on 21 datasets. People can easily reproduce results supported algorithms major via minimal efforts, such as using only one command line. In addition, provides an ecosystem deep learning architectures tools researchers to facilitate their own model development. open-source Python package that welcomes all kinds contributions.

Language: Английский

Citations

9