Bibubbles in Computational Pangenomics

• Bibubbles are minimal, bubble-like subgraphs defined on bidirected gene graphs that encode both alternative traversal paths and gene orientation variations.
• Their detection employs cycle equivalence and breadth-first search techniques, achieving computational efficiency from O(b·m²) to O(n+m) for large-scale pangenomic graphs.
• Incorporated within the SPQR-tree framework, bibubbles unify the identification of superbubbles, snarls, and ultrabubbles, thereby facilitating systematic genomic variation analysis.

Bibubbles are minimal, bubble-like subgraphs that generalize the concept of “bubble” motifs to bidirected graphs, with primary applications in computational pangenomics. Bibubbles capture all loci of local gene content or gene order variation across genomes by encoding both alternative traversal paths and bidirectionality. Their formal definition, algorithmic detection, and theoretical context are deeply tied to graph-theoretic generalizations of superbubbles, snarls, and ultrabubbles. The concept enables systematic identification and analysis of complex genomic variation, including copy-number changes, small inversions, and orientation flips, all within a unified and efficiently computable framework (Li et al., 2024, Sena et al., 9 Apr 2026).

1. Formal Definition and Graph-Theoretic Properties

Bibubbles are defined on bidirected gene graphs $G_B=(V,E)$, where vertices $V$ correspond to protein-coding genes and edges $E$ encode adjacency relationships observed on one or more genomes. Each gene $v \in V$ has two oriented endpoints, $>v$ and $<v$, representing forward and reverse orientation respectively.

A generalized bibubble is specified by a pair $(x,y)$ of oriented nodes and is minimal if it satisfies:

1. Symmetry of Reachability: $U(x,y) = U(\bar{y},\bar{x}) \neq \emptyset$, where $U(x,y)$ is the set of genes reachable from $x$ to $V$0 without passing through $V$1, $V$2 or $V$3.
2. Internal Traversability: Every $V$4 lies on some path from $V$5 to $V$6.
3. Minimality: No intermediate oriented node $V$7 yields a subpair $V$8 or $V$9 meeting criteria (1) and (2).

This definition guarantees that bibubbles are both minimal (contain no smaller such substructure) and symmetric with respect to orientation, accommodating inversions and strand-flips not captured by classic acyclic bubble motifs (Li et al., 2024).

2. Algorithmic Detection and Computational Complexity

Bibubble detection in pangenome-scale gene graphs builds on two steps:

• Cycle Equivalence: Build a contracted "net graph" by identifying cycle-equivalence classes of gene graph edges via Johnson’s linear-time algorithm (1994), enabling partitioning of the graph into locally cyclic components.
• Breadth-First Search (BFS): For each oriented node $E$0 of degree $E$1, perform BFS to collect candidate endpoints $E$2 (up to a user-set limit $E$3), where $E$4 and $E$5 share a cycle class. Each $E$6 candidate is checked for bibubble conditions above via reachability computation and minimality testing.

This strategy achieves worst-case time complexity $E$7 for $E$8 branching nodes and $E$9-capped BFS per node, with practical runtimes for human-scale graphs on the order of seconds (Li et al., 2024).

A broader unification is provided by the SPQR-tree linear-time framework (Sena et al., 9 Apr 2026), which generalizes bubble-finding for all bubble-like subgraphs:

• Construct the SPQR-tree on the underlying undirected graph, so every potential bibubble boundary is a 2-separator.
• Use dynamic programming traversals (bottom-up and top-down) on the tree to propagate “acyclicity,” “tiplessness,” and bidirected reachability states.
• Linear-time feedback arc computation enables efficient acyclicity checks for all relevant subgraphs.

This approach yields a single $v \in V$0-time algorithm capable of discovering superbubbles (directed, acyclic), snarls (bidirected, separable), ultrabubbles (snarls + tipless + acyclic), and bibubbles (custom combinations) simultaneously.

3. Relation to Other Bubble-Like Structures

A key table situates bibubbles within the spectrum of bubble-like graph structures:

Structure	Graph Type	Cycles Allowed	Inversion Capture	Minimality
Superbubble	Directed, acyclic	No	No	Yes
Weak superbubble	Directed (cyclic)	Yes	No	Yes
Snarl	Bidirected	Yes	Yes	Not required
Ultrabubble	Bidirected	Yes	Yes	Yes
Bibubble	Bidirected	Yes	Yes	Yes

Superbubbles enforce acyclic interiors in directed graphs, precluding inversion detection. Snarls loosen directionality but do not require the reachability or minimality constraints bibubbles impose. Bibubbles—by integrating orientation, reachability, cycles, and minimality—form the most general and expressive paradigm for capturing genome variation (Li et al., 2024, Sena et al., 9 Apr 2026).

4. Toy Examples and Biological Semantics

Two canonical gene-graph examples:

• Copy-number bibubble: For haplotype 1, gene order $v \in V$1–$v \in V$2–$v \in V$3; for haplotype 2, $v \in V$4–$v \in V$5–$v \in V$6. The bibubble $v \in V$7 encloses two interior vertices ($v \in V$8, $v \in V$9), reflecting a presence/absence or paralog variation.
• Inversion bibubble: One haplotype with $>v$0–$>v$1–$>v$2, another with $>v$3–(inverted $>v$4–$>v$5). The bibubble $>v$6 captures the segment inversion, inherently encoding orientation.

Biologically, bibubbles demarcate all regions where gene content and order vary—copy-number changes, inversions, duplications, and orientation switches—typically enriched in immune and fast-evolving loci. In human pangenome studies, bibubbles recover canonical polymorphic regions such as CYP2D6, C4, MAPT, HLA, and KIR gene clusters. In pan-primate analyses, breaking and merging of bibubbles can resolve evolutionary rearrangements between primate lineages (Li et al., 2024).

5. Unified Detection via SPQR-Tree Framework

The $>v$7-tree framework (Sena et al., 9 Apr 2026) identifies all bubble-like subgraphs, including bibubbles, snarls, ultrabubbles, and superbubbles, by global enumeration of 2-separators and local property checks via dynamic programming.

• At each virtual edge (representing a 2-separator), “down” and “up” states propagate minimality, acyclicity, tiplessness, and orientation-reachable conditions through the SPQR-tree.
• Linear-time feedback arc enumeration for tipless bidirected graphs enables fast acyclicity checks for potential bibubble interiors.
• Because only $>v$8 2-separators (virtual edges) are possible, the entire process (pairwise separation, state evaluation, bubble-like motif reporting) executes in overall $>v$9 time for input graphs of $<v$0 vertices and $<v$1 edges.

This result resolves the outstanding problem of efficiently listing all snarls and ultrabubbles and demonstrates that bibubble detection scales linearly for even the largest pangenomic graphs.

6. Impact and Case Studies in Pangenomics

Pangene-based bibubble detection has yielded key results on diverse genomic datasets:

• In two-genome comparisons (GRCh38 vs. T2T-CHM13), 91 bibubbles ($<v$2100 genes) encompass 691 variant genes, including PDPR, SMN2, ORM1, CCL4, NCF1, amylase, and more.
• Across 100 human haplotypes, bibubble analysis recovers 163 robust loci after conservative edge and exon filtering, revealing known and novel presence/absence assays and copy-number clusters.
• In great ape comparisons, merged gene graphs display 239 bibubbles, 131 of which remain polymorphic within humans. Evolutionary rearrangements—such as 17q21.31 inversion states and LRRC37 family duplications—are dissected via bibubble partitioning.

These findings confirm that bibubbles serve as exhaustive and theory-consistent indices for all local gene-content and gene-order variation within—and across—highly recombinogenic eukaryotic pangenomes (Li et al., 2024).

7. Theoretical Significance and Broader Implications

Bibubbles unify and extend the graph-theoretic toolkit for genome-variation analysis, providing the first comprehensive, minimal, and orientation-aware motif for all types of local sequence diversity in bidirected settings. The SPQR-tree framework ensures that bibubble detection constitutes a tractable, scalable operation, facilitating future population-scale pangenome analytics, comparative genomics, and the exploration of structural variation.

Their design reflects an overview of connectivity, reachability, acyclicity, and minimality—allowing for systematic enumeration, visualization, and downstream analysis of all nontrivial gene-variation loci. This capability underpins ongoing efforts in computational biology to transition from linear-reference genomics to fully graph-based and variation-resolved bioinformatics (Sena et al., 9 Apr 2026, Li et al., 2024).