The Omic Image Format (OIF): A Unified Machine Learning–Native Representation for Genomic and Epigenomic Data

The function of a gene product is determined by two key layers of biological information: the genetic layer, encoded by the DNA molecule, and the epigenetic layer, which regulates the timing, location, and level of gene expression. DNA sequencing and methylation data provide complementary information in classifying disease onset, progression, and phenotypic variability. However, genomic and epigenomic data formats are not natively structured for direct use in deep learning, and there is no single representation which natively combines both genomic and epigenomic data in a biologically coherent and structured way. We present the Omic Image Format (OIF), a unified, machine-learning–native representation for genomic and epigenomic data, which encodes both nucleotide sequence and DNA methylation state into a spatially coherent image with associated metadata. To enable the adoption of the OIF, we developed and publicly released OmicImage (Abraham, 2025). This open-source, workflow-agnostic Python package converts coordinate-sorted SAM alignments and optional CpG methylation reports into OIF. Using the Komagataella phaffii reference genome, the OIF preserved 100% of the positional and compositional values across both genomic and epigenomic layers and over 99% agreement in methylation values after rounding tolerance. By aligning biological features with ML architecture requirements, the OIF and the OmicImage package provide a scalable, reproducible framework for multi-omic integration that has the potential to accelerate research, improve diagnostic workflows, and advance the integration of genomics and epigenomics into precision medicine.