The Regulatory Manifold: A Coherence-Based Framework for Non-Coding Variant Analysis

The non-coding genome is increasingly recognized as a structured control layer for gene regulation rather than an inert backdrop to protein-coding exons, yet most variant interpretation methods still operate on local sequence motifs. This work proposes a coherence-based framework in which a genomic locus is modeled as a resonant lattice with boundary conditions, and gene expression emerges from the interaction of three coupled scales: domain-level chromatin topology (Mode A), intra-domain enhancer–promoter wiring (Mode B), and local transcriptional bursting (Mode C). Leveraging long-context genomic foundation models such as AlphaGenome, we interpret latent embeddings not as opaque “features” but as state vectors on a low-dimensional Regulatory Manifold. Within this manifold, single-nucleotide variants are modeled as displacement vectors: benign polymorphisms correspond to small or tangential shifts that remain within a homeostatic basin, while pathogenic variants drive the system across separatrices into mis-regulatory basins. As a concrete testbed, we outline how this framework can be applied to non-coding variants in the β-globin (HBB) locus underlying β-thalassemia, and we derive testable predictions: clinical severity should correlate more strongly with displacement in regulatory state space than with traditional local scores (e.g. conservation or motif disruption alone). This geometry-first perspective reframes non-coding variant interpretation as a signal processing problem on a learned manifold and suggests a route to unifying structure, regulation, and phenotype.