Pathotypr Brings Alignment-Free Tuberculosis Typing Into Routine Surveillance
Whole-genome sequencing has become central to tuberculosis surveillance, but the value of sequencing rises or falls on how reliably a laboratory can classify lineages, detect resistance, and compare results across programs. Pathotypr enters that gap with an alignment-free design built to work on either assembled genomes or raw short reads. In practical terms, it aims to reduce the friction between sequencing output and public health action.
The need for a broader lineage framework is obvious from the structure of the Mycobacterium tuberculosis complex itself. Ten human-adapted lineages, L1 through L10, sit alongside four animal-adapted clades, A1 through A4, yet several uncommon or newly described groups have remained poorly represented in existing typing schemes. When a lineage system does not include that diversity, an isolate can be left unclassified or pushed toward the nearest available label. In surveillance, a wrong label is not a cosmetic problem; it can distort transmission mapping, obscure the geographic spread of rare lineages, and weaken comparisons between regions.
Pathotypr was built as open-source software with three main outputs: lineage assignment, World Health Organization-cataloged resistance-associated variants, and a closely related reference genome for downstream work. For assemblies, the platform can classify genomes through phylogenetic markers or through an independent k-mer model. For reads, it screens sequencing data directly for lineage and resistance markers and can return more than one supported lineage when mixed signals appear. Mixed signals matter in tuberculosis because contaminated samples, mixed infections, or imperfect assemblies can all produce misleading results if the software forces a single answer too early.
A 26,813-Genome Phylogeny Anchored the Marker Set
The lineage framework behind Pathotypr came from a maximum-likelihood phylogeny of 26,813 MTBC genomes and 609,003 polymorphic sites. That tree placed L1 through L10 and A1 through A4 within a single evolutionary backbone and made the underrepresentation of several lineages impossible to ignore. The developers then traced substitutions onto branches and selected single-nucleotide polymorphisms on the stem branch leading to each monophyletic lineage or clade.
After filtering out 3,039 missense variants and 454 homoplastic nonmissense variants, 2,145 of 5,638 candidates remained as lineage-defining markers. Those numbers show a deliberately conservative approach. By removing variants more likely to confuse classification, the marker set prioritizes specificity over a bloated inventory of polymorphisms that may not hold across global isolates. The resulting marker composition represented 33% to 46% of total SNPs across lineages, a range that supports the filtering strategy without pretending every lineage is equally sampled.
A striking point sits beneath those totals. The smallest lineages, including L7 through L10 and A1 through A3, may still harbor diversity not captured in the current dataset. Sparse sampling is a real constraint in tuberculosis genomics, because a classification tool can only be as complete as the evolutionary breadth it has seen. Rare lineages are exactly where public health systems need help, yet they are also the groups most likely to be missed.
Concordance Expanded Across a Very Large Benchmark
Pathotypr was evaluated in independent datasets, an 88,071-sample benchmark, and 7,148 genomes from the CRyPTIC dataset. Across those evaluations, the tool improved concordance, expanded uncommon lineage coverage, and detected resistance markers. The scale of the benchmark is important because tuberculosis surveillance rarely operates in tidy, small, curated sets. Real-world pipelines must cope with heterogeneous sequencing quality, incomplete typing frameworks, and isolates that fall outside the best-characterized lineages.
Comparisons with existing tools showed variable marker overlap. Pathotypr shared no lineage markers with some tools and as many as 698 markers, or 62.8%, with TB-Profiler. Overlap with SNP-IT reached 1,750 markers, or 12.9%. Those figures are not a contest for its own sake. They show that different surveillance tools are not drawing from identical marker inventories, which helps explain why misclassification can persist even when laboratories believe they are using similar genomic logic.
The software’s ability to operate on both assemblies and reads is especially useful in tuberculosis control, where turnaround time and data format often differ from one setting to the next. Assemblies are useful for downstream phylogenetics and comparative analysis, but raw reads are often available sooner and are less dependent on assembly quality. By accepting either input, Pathotypr gives public health labs a path from sequencing to lineage and resistance calls without forcing every workflow through the same analytic bottleneck.
Resistance Detection and Reference Matching Add Public Health Value
Resistance detection is where the tool moves beyond classification. Pathotypr screens for World Health Organization-cataloged resistance-associated variants, which gives surveillance teams a way to connect lineage assignment with treatment-relevant genomic information. In tuberculosis, those two outputs belong together. A lineage call can help explain outbreak structure and spread, while resistance markers shape regimen choice, infection control, and the interpretation of persistent transmission.
The platform also identifies the closest reference genome using k-mer containment. That feature should make comparative analysis more efficient because it can point investigators toward a reference genome that better matches the isolate under study. For highly diverse or uncommon lineages, choosing an appropriate reference is not a trivial detail. A distant reference can make variant interpretation more cumbersome and can subtly skew how sequence differences are framed in follow-up analyses.
By design, Pathotypr appears aimed at the work that follows the initial sequencing run. It is not just a classifier; it is a bridge between read-level data, lineage interpretation, and resistance surveillance. In a field where whole-genome sequencing is increasingly used to distinguish local transmission from repeated introduction, that combination is operationally useful.
Sampling Gaps Still Shape What the Tool Can Claim
The study also leaves one caution plainly visible. The lineages with the fewest genomes, especially L7 through L10 and A1 through A3, may not be fully represented in the marker set because unsampled diversity can remain hidden for years. In surveillance systems, this kind of underrepresentation tends to affect the least common populations first, which is exactly where genomic tools are often most needed.
Pathotypr therefore expands the practical reach of tuberculosis genomic surveillance, but it does not erase the limits imposed by sampling imbalance. Its marker set is broad, its input flexibility is useful, and its resistance calling adds clinical relevance, yet the accuracy of future classifications will still depend on how well the global MTBC diversity is captured in reference datasets. The next question for public health teams is not whether alignment-free typing can help. It is how quickly surveillance networks can incorporate rare lineages, mixed samples, and resistance patterns into routine use without losing comparability across programs.