Correcting gradient-based interpretations of deep neural networks for genomics

Majdandzic, Antonio, Rajesh, Chandana, Koo, Peter K (May 2023) Correcting gradient-based interpretations of deep neural networks for genomics. Genome Biology, 24 (1). p. 109. ISSN 1474-760X

PDF Correcting gradient-based interpretations of deep neural networks for genomics.pdf - Published Version Available under License Creative Commons Attribution. Download (2MB)

Abstract

Post hoc attribution methods can provide insights into the learned patterns from deep neural networks (DNNs) trained on high-throughput functional genomics data. However, in practice, their resultant attribution maps can be challenging to interpret due to spurious importance scores for seemingly arbitrary nucleotides. Here, we identify a previously overlooked attribution noise source that arises from how DNNs handle one-hot encoded DNA. We demonstrate this noise is pervasive across various genomic DNNs and introduce a statistical correction that effectively reduces it, leading to more reliable attribution maps. Our approach represents a promising step towards gaining meaningful insights from DNNs in regulatory genomics.

Item Type:	Paper
Subjects:	bioinformatics bioinformatics > genomics and proteomics organs, tissues, organelles, cell types and functions > tissues types and functions > neural networks organs, tissues, organelles, cell types and functions organs, tissues, organelles, cell types and functions > tissues types and functions
CSHL Authors:	Majdandzic, Antonio Koo, Peter K Rajesh, Chandana
Communities:	CSHL labs > Koo Lab CSHL Cancer Center Program CSHL Cancer Center Program > Gene Regulation and Inheritance Program
SWORD Depositor:	CSHL Elements
Depositing User:	CSHL Elements
Date:	9 May 2023
Date Deposited:	28 Sep 2023 19:25
Last Modified:	09 Feb 2024 15:28
PMCID:	PMC10169356
Related URLs:	Publisher
URI:	https://repository.cshl.edu/id/eprint/41046

Actions (login required)

Administrator's edit/view item