Locality-Controlled Out-of-Distribution Guidance for Robust Regulatory DNA Sequence Design
Abstract
Designing regulatory DNA sequences for targeted gene-expression control is an important challenge. Recent approaches frame this task as property-conditioned generation, where diffusion models trained on natural sequences are guided toward high predicted activity through reward-based fine-tuning or sampling-time guidance. We show that single-reward evaluation can obscure reward-hacking behavior. By decomposing evaluation into property, novelty, and plausibility, we identify an empirically underexplored region of the evaluation space in which all three objectives are satisfied, but no existing baseline succeeds. We postulate that this gap arises because prior methods treat guidance primarily as a matter of magnitude. In regulatory DNA, however, sequence grammar is spatially nonuniform: motif positions are sensitive to perturbation, while background regions allow more exploration. We therefore propose LOCOGEN by reformulating guidance as a problem of locality. LOCOGEN learns an invariant motif mask through sequence-level invariance learning to protect motif positions, while applying OOD damping only to background positions. It operates at sampling time on a frozen diffusion backbone, without retraining the backbone. Across HepG2, K562, and SK-N-SH, LOCOGEN reaches this previously underexplored region of the property-novelty-plausibility space. Code is available at https://github.com/Lotfollahi-lab/LocoGen.
Related
- Evidence for a regulatory role of the evolutionarily conserved sequence complementarity between tRNAs
- Few-shot regulatory sequence design with a genomic language model
- Multi-model biological and sequence information fusion for gene regulatory network inference from single-cell transcriptomics
- Decoding super-enhancer regulatory grammar and pathogenic gain-of-function variants directly from DNA sequence
- DeepSCENIC: transfer learning from sequence-to-function models enables causal gene regulatory network inference
BibTeX
@misc{jeong2026localitycontrolled,
title = {{Locality-Controlled Out-of-Distribution Guidance for Robust Regulatory DNA Sequence Design}},
author = {Dabin Jeong and Marwane Bourdim and Mo Lotfollahi},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.41736},
eprint = {2610.41736},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.41736},
note = {CC BY-NC-SA 4.0},
}
RIS
TY - UNPB
TI - Locality-Controlled Out-of-Distribution Guidance for Robust Regulatory DNA Sequence Design
AU - Dabin Jeong
AU - Marwane Bourdim
AU - Mo Lotfollahi
PY - 2026
DA - 2026/10/30
PB - arXiv
DO - 10.48550/arxiv.2610.41736
UR - https://arxiv.org/abs/2610.41736
AB - Designing regulatory DNA sequences for targeted gene-expression control is an important challenge. Recent approaches frame this task as property-conditioned generation, where diffusion models trained on natural sequences are guided toward high predicted activity through reward-based fine-tuning or sampling-time guidance. We show that single-reward evaluation can obscure reward-hacking behavior. By decomposing evaluation into property, novelty, and plausibility, we identify an empirically underexplored region of the evaluation space in which all three objectives are satisfied, but no existing baseline succeeds. We postulate that this gap arises because prior methods treat guidance primarily as a matter of magnitude. In regulatory DNA, however, sequence grammar is spatially nonuniform: motif positions are sensitive to perturbation, while background regions allow more exploration. We therefore propose LOCOGEN by reformulating guidance as a problem of locality. LOCOGEN learns an invariant motif mask through sequence-level invariance learning to protect motif positions, while applying OOD damping only to background positions. It operates at sampling time on a frozen diffusion backbone, without retraining the backbone. Across HepG2, K562, and SK-N-SH, LOCOGEN reaches this previously underexplored region of the property-novelty-plausibility space. Code is available at https://github.com/Lotfollahi-lab/LocoGen.
ER -