Perfect Phylogeny Haplotyping is Complete for Logspace

Haplotyping is the bioinformatics problem of predicting likely haplotypes based on given genotypes. It can be approached using Gusfield’s perfect phylogeny haplotyping (PPH) method for which polynomial and linear time algorithms exist. These algorithm use sophisticated data structures or do a stepwise transformation of the genotype data into haplotype data and, therefore, need a linear amount of space. We are interested in the exact computational complexity of PPH and show that it can be solved space-efficiently by an algorithm that needs only a logarithmic amount of space. Together with the recently proved L-hardness of PPH, we establish L-completeness. Our algorithm relies on a new characterization for PPH in terms of bipartite graphs, which can be used both to decide and construct perfect phylogenies for genotypes efficiently.

💡 Research Summary

The paper investigates the exact computational complexity of the Perfect Phylogeny Haplotyping (PPH) problem, a central task in bioinformatics where one must infer a set of haplotypes that explain a given genotype matrix while forming a perfect phylogeny. Historically, PPH has been tackled by polynomial‑time algorithms, most notably Gusfield’s linear‑time method, which rely on sophisticated data structures and require linear space because they maintain the entire genotype matrix and auxiliary structures throughout execution.

The authors’ primary contribution is a new structural characterization of PPH in terms of bipartite graphs. They construct a graph whose vertices correspond to alleles (the two possible values at each genotype site) and whose edges connect alleles that appear together in a genotype but belong to opposite partitions. They prove that the genotype matrix admits a perfect phylogeny if and only if this graph is bipartite. This equivalence collapses the previously intricate combinatorial conditions into a single graph‑theoretic property, thereby opening the door to space‑efficient algorithms.

Building on this insight, the paper presents a logspace algorithm for deciding PPH. The algorithm processes the input genotype matrix in a streaming fashion, using only O(log n) bits of working memory, where n is the size of the input. It performs a depth‑first‑search‑like traversal of the bipartite graph without an explicit stack: the current vertex’s colour (partition) is inferred from previously visited vertices, and a small counter suffices to backtrack when necessary. Because bipartiteness can be checked by attempting a 2‑colouring, the algorithm merely needs to store the colour of the current vertex and a few auxiliary flags, all of which fit within logarithmic space.

When the graph is confirmed to be bipartite, the same logspace framework is used to construct an explicit set of haplotypes. The colour assignment directly translates into a haplotype configuration: each vertex’s side of the bipartition determines which allele appears in each haplotype. No additional memory beyond the decision phase is required, so both decision and construction are achieved within the same space bound.

The authors also reference a recent L‑hardness result for PPH, which shows that the problem is at least as hard as any problem in deterministic logspace. By providing a deterministic logspace algorithm, they close the gap and establish that PPH is L‑complete. This places PPH among the most fundamental problems that can be solved with only logarithmic workspace, a classification that had not been previously known for a biologically motivated combinatorial problem.

From a practical perspective, a logspace solution is highly attractive for environments where memory is scarce, such as mobile devices, embedded sequencing platforms, or large‑scale streaming pipelines that cannot afford to keep the entire genotype matrix in RAM. The algorithm’s streaming nature also reduces I/O overhead and can be combined with external‑memory techniques for even larger datasets.

The paper concludes with several avenues for future work. Extending the bipartite‑graph characterization to multi‑allelic loci or to models that tolerate genotyping errors could broaden the applicability of the logspace approach. Investigating whether similar graph‑based reductions exist for related phylogeny problems (e.g., near‑perfect phylogenies, incomplete lineage sorting) may yield further low‑space algorithms. Finally, the authors suggest exploring parallel or distributed implementations that preserve the logarithmic space guarantee while achieving higher throughput.

In summary, the work delivers a clean theoretical breakthrough—showing that Perfect Phylogeny Haplotyping is L‑complete—by introducing a bipartite‑graph characterization and a deterministic logarithmic‑space algorithm for both decision and construction. This advances our understanding of the problem’s inherent difficulty and opens practical pathways for memory‑constrained haplotype inference.