Uncovering BioLOGICAL Motifs and Syntax via Sufficient and Necessary Explanations
Keywords: interpretability, attributions, computational biology
TL;DR: This paper proposes a new interpretability method for genomic deep neural networks, based off sufficient and necessary, that is able to detect motifs and the syntax that drives transcription factor binding.
Abstract: In recent years, deep neural networks (DNNs) have excelled at learning from high-throughput genome-profiling experiments to predict transcription factor (TF) binding. TF binding is driven by sequence motifs, and explaining how and why DNNs make accurate predictions could help identify these motifs, as well as their logical syntax. However, the black-box nature of DNNs makes interpretation difficult. Most post-hoc methods evaluate the importance of each base pair in isolation, often resulting in noise since they overlook the fact that motifs are contiguous regions. Additionally, these methods fail to capture the complex interactions between different motifs. To address these challenges, we propose Motif Explainer Models (MEMs), a novel explanation method that uses sufficiency and necessity to identify important motifs and their syntax. MEMs excel at identifying multiple disjoint motifs across DNA sequences, overcoming limitations of existing methods. Moreover, by accurately pinpointing sufficient and necessary motifs, MEMs can reveal the logical syntax that governs genomic regulation.
Supplementary Material: zip
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 12933
Loading