PAUSE: principled feature attribution for unsupervised gene expression analysis

Genome Biology

Method

John C. Russell¹  Ting-I Lee¹  Ben W. Blue¹  Matt Kaeberlin²  Anna Spiro³  Su-In Lee³  Joseph D. Janizek⁴  Safiye Celik⁵ 

[1]Department of Pathology, University of Washington, Seattle, USA

[2]Department of Pathology, University of Washington, Seattle, USA

[3]Department of Genome Sciences, University of Washington, Seattle, USA

[4]Paul G. Allen School of Computer Science and Engineering, University of Washington, Seattle, USA

[5]Paul G. Allen School of Computer Science and Engineering, University of Washington, Seattle, USA

[6]Medical Scientist Training Program, University of Washington, Seattle, USA

[7]Recursion Pharmaceuticals, Salt Lake City, USA

关键词: Transcriptomics;    Gene expression;    Deep learning;    Explainable AI;    Unsupervised learning;    Feature attribution;   

DOI  :  10.1186/s13059-023-02901-4

 received in 2022-04-30, accepted in 2023-03-17,  发布年份 2023

来源: Springer

【 摘 要 】

As interest in using unsupervised deep learning models to analyze gene expression data has grown, an increasing number of methods have been developed to make these models more interpretable. These methods can be separated into two groups: post hoc analyses of black box models through feature attribution methods and approaches to build inherently interpretable models through biologically-constrained architectures. We argue that these approaches are not mutually exclusive, but can in fact be usefully combined. We propose PAUSE (https://github.com/suinleelab/PAUSE), an unsupervised pathway attribution method that identifies major sources of transcriptomic variation when combined with biologically-constrained neural network models.

【 授权许可】

CC BY   
© The Author(s) 2023

【 图 表 】

Fig. 5

Fig. 4

40507_2023_167_Article_IEq80.gif

Fig. 14

Fig. 13

Fig. 12

Fig. 11

Fig. 4

40507_2023_167_Article_IEq69.gif

Fig. 6

Fig. 5

Fig. 1

40507_2023_167_Article_IEq60.gif

Fig. 6

Fig. 4

40507_2023_167_Article_IEq57.gif

Fig. 2

40507_2023_167_Article_IEq56.gif

Fig. 2

40507_2023_167_Article_IEq53.gif

40507_2023_167_Article_IEq52.gif

Fig. 1

Fig. 3

Fig. 3

Fig. 2

Fig. 2

Fig. 2

40507_2023_167_Article_IEq44.gif

Fig. 1

Fig. 2

Fig. 1

Fig. 2

Fig. 4

40507_2023_167_Article_IEq38.gif

Fig. 1

12942_2022_316_Article_IEq23.gif

Fig. 6

Fig. 1

10194_2023_1579_Article_IEq40.gif

Fig. 2

Fig. 2

10194_2023_1579_Article_IEq37.gif

10194_2023_1579_Article_IEq36.gif

Fig. 1

12942_2022_316_Article_IEq12.gif

10194_2023_1579_Article_IEq31.gif

10194_2023_1579_Article_IEq29.gif

Fig. 1

12942_2022_316_Article_IEq4.gif

Fig. 5

10194_2023_1579_Article_IEq24.gif

10194_2023_1579_Article_IEq22.gif

10194_2023_1579_Article_IEq18.gif

Fig. 5

Fig. 8

10194_2023_1579_Article_IEq13.gif

Fig. 7

Fig. 11

Fig. 2

Fig. 4

Fig. 10

Fig. 1

Fig. 14

Fig. 1

Fig. 8

Fig. 1

Fig. 1

Fig. 9