Learning Residual Set Representations for Mechanistic Gene Regulation

Confirmation: I have read and agree with the workshop's policy on behalf of myself and my co-authors.

Track: long paper (4–8 pages excluding references)

Keywords: Gene regulatory networks, combinatorial gene regulation, exact regulator set recovery, set-based representation learning, residual high-order interactions, single-cell transcriptomics, mechanistic interpretability

Abstract:

Learning biologically meaningful representations requires alignment between model outputs and
underlying regulatory mechanisms. In gene regulatory networks (GRNs), transcriptional regulation is
inherently combinatorial: target genes are controlled by specific sets of interacting regulators. However,
most machine learning approaches rely on edge-level representations that fail to capture this regulatory
logic and are evaluated using pairwise objectives.
We formulate target-specific GRN inference as Exact Regulator Set Recovery and introduce
a two-stage framework that separates candidate retrieval from set-level reasoning. Stage-1 retrieves a
high-recall candidate pool using a target-conditioned attention retriever. Stage-2 selects the regulator
team using Residual HOS2, which models non-additive interactions as a correction to decomposable
pairwise evidence. Experiments on single-cell data with known combinatorial ground truth show that this
residual set formulation enables more stable learning and improves recovery of mechanistically meaningful
regulator teams in sparse genomic regimes.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Presenter: ~Siddhi_Kanta_Mishra1

Format: Yes, the presenting author will attend in person if this work is accepted to the workshop.

Funding: Yes, the presenting author of this submission falls under ICLR’s funding aims, and funding would significantly impact their ability to attend the workshop in person.

Submission Number: 77