Work

Computable Genomes: A Layered Computational Framework for Non-Coding DNA and Cellular Behavior

Year: 2026
Type: preprint
Abstract: Genome-wide association studies have catalogued tens of thousands of trait-associated loci, most of them in non-coding DNA, yet for most loci the path from nucleotide to phenotype remains correlational. We propose a layered computational decomposition of the genome as a complement to data-driven approaches: sequence as formal language, sequence as information-bearing signal, mobile elements as runtime, compilation to interaction networks through thermodynamic and rule-based models, and execution as cellular and multicellular dynamics. Each layer is defined by its inputs, outputs and formalisms, and we set out where the frame breaks down, including the looseness of algorithmic-information estimates, the triviality of universality results and the failure of string grammars at the chromatin level. Synthetic-biology platforms serve as tests of whether the layers compose, and genomic foundation models and the AlphaGenome Atlas, which predicts effects for all nine billion possible single-nucleotide variants, serve as oracles whose coverage and limits we map onto the framework. We state five falsifiable predictions and test the first, that sequence compressibility carries information about regulatory function beyond composition and conservation, on 38,556 single-nucleotide variants from a saturation-mutagenesis reporter assay of 21 regulatory elements. In a pooled model, local compressibility is associated with measured effect size (standardized β = −0.030, p = 0.0014). The association roughly halves under full control for base composition, loses significance when composition, motif and noise adjustments are combined under locus-level inference (wild cluster bootstrap p = 0.11), and averages zero within loci (random-effects β = −0.0003, p = 0.90). The prediction is therefore not supported at the variant level in these data, which bound the effect a genome-wide test should expect. The argument is not that the genome "is" a program, but that treating it as one supplies formal tools whose predictions can be checked.
Language: English
Open Access status: green
OpenAlex ID: W7235992503
Cites: 92
Cited by:
Related to: 10