Geometry is All You Need

Aging is written not only in our DNA sequence but also in the geometry of chromatin. Circulating cell-free DNA (cfDNA) fragments preserve nucleosomal structure, rotational phasing, and patterns of disorder, yet current diagnostic approaches, including DNA methylation clocks, ignore this rich information layer. Here we introduce a new representational framework for cfDNA, defining geometry tokens (fragment length, 10 bp rotational phase, sequence motifs, GC content, regional bins) and graph edges (linear adjacency, phase similarity, loop priors). These features are embedded and processed by a Geometry-Aware Transformer (GAT™), an architecture that integrates local graph attention with global transformer encoding. The GAT is trained with self-supervised objectives, masked feature recovery, edge prediction, contrastive pre/post learning, and coverage reconstruction, alongside supervised heads that predict biological age, delta-age, inflammation, and disease risk. This framework establishes chromatin geometry fragmentomics as a new scientific domain. It provides a scalable, interpretable, and clinically actionable platform for measuring biological aging. To our knowledge, this is the first demonstration that cfDNA chromatin geometry can be represented as a tokenised language and analysed with transformer-based architectures, establishing a new field of chromatin geometry fragmentomics.