An Allele-Centric Pan-Graph-Matrix Representation for Scalable Pangenome Analysis

Population-scale pangenome analysis increasingly requires representations that unify single-nucleotide and structural variation while remaining scalable across large cohorts. Existing formats are typically sequence-centric, path-centric, or sample-centric, and often obscure population structure or fail to exploit carrier sparsity. We introduce the H1 pan-graph-matrix, an allele-centric representation that encodes exact haplotype membership using adaptive per-allele compression. By treating alleles as first-class objects and selecting optimal encodings based on carrier distribution, H1 achieves near-optimal storage across both common and rare variants. We further introduce H2, a path-centric dual representation derived from the same underlying allele-haplotype incidence information that restores explicit haplotype ordering while remaining exactly equivalent in information content. Using real human genome data, we show that this representation yields substantial compression gains, particularly for structural variants, while remaining equivalent in information content to pangenome graphs. H1 provides a unified, population-aware foundation for scalable pangenome analysis and downstream applications such as rare-variant interpretation and drug discovery.

💡 Research Summary

The paper addresses a fundamental limitation of current genomic data formats for population‑scale pangenome analysis. Traditional formats such as VCF/BCF are sample‑centric: they organize variants by genomic position and store genotypes per sample in a fixed layout. While this works well for per‑sample queries, it obscures the population‑level question “which haplotypes carry a given allele”. Graph‑based pangenome representations (nodes, edges, and implicit haplotype paths) make structural variation explicit but still hide the allele‑to‑haplotype incidence relation, and they are primarily driven by sequence redundancy rather than carrier sparsity.

To overcome these issues, the authors introduce the H1 pan‑graph‑matrix, an allele‑centric binary matrix where each row corresponds to a concrete allele (single‑nucleotide, indel, or structural variant) and each column corresponds to a haplotype. The key innovation is per‑allele adaptive encoding: for a given allele with carrier count k among H haplotypes, the representation chooses either a dense bitmap (cost ≈ H bits) or a sparse list of carrier identifiers (cost ≈ k·⌈log₂ H⌉ bits). By equating the two costs they derive a break‑even threshold k⁎ ≈ H·log₂ H. Alleles with k ≪ k⁎ are stored as sparse lists, while common alleles with k ≫ k⁎ use dense bitmaps. This simple cost model is dataset‑independent; it follows from the structure of the encoding family rather than implementation tricks. Consequently, H1 automatically forms a hybrid “dense‑sparse” matrix that closely tracks the lower envelope of both encoding schemes across the entire allele‑frequency spectrum.

The authors also formalize the mathematical relationship between H1 and pangenome graphs. The matrix can be interpreted as the incidence algebra of the graph: each bubble (alternative path) in the graph maps to one or more rows, each haplotype path maps to a column, and a non‑zero entry indicates that the haplotype traverses the allele’s branch. This correspondence is exact, guaranteeing that no information is lost when moving between the graph and matrix views.

Because many analyses require explicit haplotype ordering (e.g., reconstructing variant sequences along a chromosome), the paper introduces H2, a path‑centric dual representation derived directly from H1. H2 stores each haplotype as an ordered list of abstract edges that correspond to reference segments and variant‑induced alternatives. An inverted index maps each edge back to the set of haplotypes that traverse it, enabling efficient adjacency and topology queries. Importantly, H1 and H2 are information‑equivalent: the carrier set for any allele in H1 exactly matches the set of haplotypes whose H2 paths include the corresponding edge, and vice versa. Thus H1 is optimized for population‑level sparsity and compression, while H2 is optimized for path‑order‑dependent analyses.

The authors evaluate the approach on real human data from the 1000 Genomes Project. They use phased SNV/INDEL and structural variant callsets for 200 diploid individuals (400 haplotypes) across a 2 Mb region on chromosome 1. The region contains 24,921 SNV/INDEL sites and 45 true structural variants after filtering. Structural variants are extremely sparse: >50 % are carried by ≤2 haplotypes and 87 % have allele frequency <10 %. Applying the adaptive hybrid encoding, H1 achieves a 78 % reduction in storage for structural variants relative to a bitmap‑only representation, and a 69 % reduction for SNVs. These gains align closely with the theoretical break‑even point, confirming that the cost model accurately predicts optimal encoding choices.

The paper also discusses graph construction strategies. A naïve graph that creates a bubble for every SNV results in tens of thousands of nodes, making visualization and downstream analysis cumbersome. Instead, the authors propose a “structural‑variant‑focused” graph where only large rearrangements form bubbles, while SNVs are stored as annotations or as side nodes attached to a coarsened backbone (e.g., segmented every 1 kb). This yields a compact graph that preserves structural context while keeping size manageable. The matrix view (H1) explains why such a graph can be compressed: the sparsity of carrier sets drives storage efficiency, whereas the graph view explains structural topology.

In summary, the paper makes five major contributions: (1) a novel allele‑centric matrix representation (H1) that treats alleles as first‑class objects; (2) a simple, dataset‑independent cost model with a derived break‑even threshold for dense vs. sparse encoding; (3) a formal proof of information‑equivalence between the matrix and pangenome graph (incidence algebra interpretation); (4) the introduction of H2, a path‑centric dual that restores explicit haplotype ordering while preserving exact information; and (5) empirical validation on human genomic data showing substantial compression, especially for rare structural variants. By separating population‑level carrier analysis (H1) from haplotype‑order analysis (H2) yet keeping them tightly linked, the framework provides a scalable foundation for large‑scale pangenome projects, rare‑variant interpretation, and downstream applications such as drug discovery.