Normalization, Not Depletion, Is the Binding Constraint in Multi-Site Wastewater Metagenomics

Question: How are computational pipelines for untargeted wastewater metagenomics handling host and bacterial read depletion, taxonomic classification, and abundance normalization across large multi-site sequencing datasets?

Answer

Across the retrieved papers, the field has quietly resolved the first two problems and stalled on the third. Host-read removal has been relegated to a generic, reproducible preprocessing step performed by fast pangenome-indexed filters rather than a bespoke wastewater stage [13] [12], and the bacterial background that dominates wastewater libraries is not depleted at all — it is out-competed by hybrid-capture or primer-spike enrichment [20] [21]. Taxonomic classification at scale rests on k-mer classifiers plus control-based filtering, with the residual error attributed explicitly to reference-database incompleteness rather than to algorithm [18] [24]. The genuinely unsettled piece is abundance normalization: the largest retrieved evaluation shows that the conventional pathogen-fraction-of-total-reads summary is poorly suited to quantitative use, and that endogenous dietary plant virus (tobamovirus) counts outperform six alternatives across 25 U.S. sites [16]. Taken together, these papers support the position that multi-site comparability in wastewater metagenomics is currently bought with enrichment and carefully chosen denominators, not with depletion or better classifiers.

Host-read removal has become a generic alignment-free preprocessing step rather than a wastewater-specific design choice

The methodological argument for host depletion in wastewater is legal as much as analytical: many jurisdictions impose strict privacy and data-protection constraints on human sequence data, so microbiome data containing partly human reads cannot easily be processed further or published [13]. Cleanifier answers this with gapped k-mers (spaced seeds) over a human pangenome index stored in a probabilistic Cuckoo filter, using a sampling mode that makes it the fastest of the compared state-of-the-art methods at comparable accuracy, with memory requirements similar to minimizer-index methods [13]. CLEAN generalizes the same slot in the pipeline: it removes technology-specific control spike-ins alongside host reads and rRNA, and its authors argue that decontamination is routinely overlooked or performed by irreproducible, untraceable steps [12]. The retrieved papers thus treat host removal as a hygiene step with a reproducibility requirement, not as a wastewater-specific modelling problem.

Bacterial reads are treated as signal and out-competed rather than subtracted

No retrieved wastewater pipeline depletes bacterial reads; instead the bacterial and eukaryotic background is suppressed by front-end enrichment. Hybrid capture applied to roughly 450 weekly samples from 15 wastewater treatment plants achieved a median viral read fraction above 19%, and the resulting relative abundances of common pathogens correlated with RT-ddPCR quantification [20]. Metagenomic sequencing with spiked primer enrichment (MSSPE), designed with a fully open-source primer-design algorithm, achieved a median two-fold enrichment of reads per million across nine pathogenic DNA and RNA viruses when explicitly addressing samples dominated by bacterial and eukaryotic nucleic acid background [21]. Where no enrichment is used, the substitute is brute sequencing depth: untargeted random-hexamer sequencing at an average of 1.1 billion reads per sample detected enteric viruses in virtually every sample as well as unexpected pathogens [22]. The scale reached by this strategy is now substantial — 1,206 samples from 27 sites in nine states covering 13 million people at approximately one billion read pairs per sample, totalling 1.2 trillion read pairs [23].

k-mer classifier errors in low-target-fraction samples are contained by shuffled-sequence controls and coverage-breadth filters, not by new classifiers

Kraken2's false positives in low-microbial-biomass samples arise from shared compact hash codes across distinct k-mers, hash collisions, and reference-database structure; in high-non-microbial-noise samples the classified taxa correlate significantly with those obtained from shuffled sequences at default settings, and a confidence threshold of at least 0.2 reduces misclassification, though the required threshold varies by taxon [18]. The same sensitivity to filtering options is documented for the Kraken family in ancient metagenomics, where the proposed remedy is depth-conditioned thresholding rather than tool substitution [17]. A structurally different fix is to use matched metagenomic coverage breadth as a prior: requiring 92–95% coverage breadth before accepting a genome reduced detected operational genomic units by 99.4% while retaining 87.8% of aligned reads, raising F1 from 0.02 to 0.61, with residual false positives attributable to genomes with extremely high nucleotide similarity to true members and false negatives enriched among low-abundance taxa [6]. On the long-read side, assembly-driven database reduction combined with expectation-maximization contig reassignment improved precision over existing tools, and read classification alone performed on par with them [1].

Resolution is being pushed to haplotype level by placing reads on phylogenies rather than assigning them to taxa

Where the target is a known pathogen lineage rather than a community, k-mer or marker assignment is bypassed: WEPP places wastewater reads onto mutation-annotated trees containing all globally available clinical sequences and their ancestral nodes, reports per-haplotype abundance, and flags unaccounted alleles as candidate novel variants, typically recovering haplotypes within one single-nucleotide mutation of truth [8]. For bacterial pathogens below the abundance needed for de novo assembly, taxonomic read binning combined with reference-based assembly reconstructed 95–99% complete genomes from reads at 0.1–1% relative abundance [25].

Unclassifiable read fractions are being addressed by building wastewater-specific reference space

The scale of the residue is documented: read-level classification leaves much of wastewater data unclassifiable, and a consortium-built wastewater virus genome database of 21,015 near-complete genomes — 79% ssRNA bacteriophage, fewer than 4,000 with matches in prior databases — captured roughly one fifth of the reads Kraken2 could not classify [24]. Parallel resources target the bacterial and functional side, including protein families from 580 metagenomes and 132 metatranscriptomes with taxonomic and structural annotation [2], and a global database built from 1,344 publicly available metagenomes across six continents with recovered bins, contigs, and reads [7]. Compute orchestration for such collections is correspondingly a first-class concern: a cloud-oriented workflow with machine-learning-tuned assembler memory allocation was demonstrated on 757 sewage metagenomes [3].

Endogenous fecal and dietary markers, not total-read fractions, carry multi-site quantitative signal

The decisive constraint is compositional: because a taxon's apparent abundance depends on every other taxon in the sample, a pathogen's fraction of total reads is poorly suited to quantitative use, and normalization by tobamovirus-genus counts — diet-derived plant viruses abundant in human stool — improved median site concordance in 18 of 19 pathogen-and-comparison-source combinations across 1,425 samples from 25 sites, with median gains roughly five times larger than median losses, reaching clinical concordance comparable to targeted wastewater PCR [16]. Gains were largest for year-round-circulating SARS-CoV-2 and norovirus and smaller for sharply seasonal influenza and RSV, where baseline concordance was already high [16]. Absolute quantification is pursued separately through spike-ins: synthetic DNA standards with entropy-based detection thresholds and mapping-error correction yielded a detection limit of roughly 500 copies/L at 200 million reads per sample, far above PCR's roughly 10 copies/L [9], while multi-species genomic spike-in calibration achieved significantly higher calibration accuracy than single-fragment or single-organism spike-ins and species-level absolute results comparable to digital PCR [11]. Statistical treatment of compositionality is also being reframed at the group level, with group-wise relative log expression and fold-truncated sum scaling holding the false discovery rate where existing normalization methods fail [15].

Conflicts and gaps

The normalization literature does not reconcile its denominators. One large multi-site evaluation ranks tobamovirus counts first among seven approaches [16], while a paired-matrix study comparing PMMoV, carjivirus, and 16S normalization concludes that using different controls together complicates interpretation of concentrations across targets [19]; no retrieved paper benchmarks tobamovirus against these markers head-to-head. Absolute spike-in approaches [9] [11] and endogenous-marker approaches [16] are never compared at multi-site scale. A second tension is between pooled normalization and site-specific characterization: limit-of-detection modelling concludes that mNGS detection performance varies with sequencing method and analysis process and that site-specific method characterization is required because sites differ in system configuration, background organisms, and inhibitors [14] — a claim in direct tension with the premise that a single denominator transfers across 25 sites. Third, bacterial read depletion as such is absent from the retrieved wastewater literature; the closest analogues are clinical host-depletion methods validated outside wastewater [10]. Fourth, the normalization schemes are blind to two documented confounders: sewer biofilm communities contribute a mean 9% (range 0.9–23.3%) of wastewater bacterial communities and carry their own pathogen and resistance gene profiles [5], and ARG associations with antibiotic use differ by putative bacterial host, with Enterobacteriaceae-associated genes correlating directly and Pseudomonadaceae-associated genes showing a lagged relationship [4].

What would change the answer

A single multi-site ring trial would settle most of the above: distribute identical mock communities plus a common quantitative spike-in panel across all participating sites, sequence them within the production workflow, and use the known input to benchmark tobamovirus, PMMoV, carjivirus, 16S, total-read, and multi-species spike-in denominators on the same samples across a full seasonal cycle — the design that would put [16], [19], [11], and [9] into one frame. Second, validate coverage-breadth filtering [6] and shuffled-sequence null controls [18] against wastewater spike-ins of known composition rather than gut or ancient mock data, which would establish whether the reported false-positive structure transfers to a matrix dominated by non-target bacterial DNA. Third, quantify the gain from wastewater-specific reference space by re-classifying a fixed, held-out sample set with and without the new databases [24] [7]. Finally, fit an explicit compositional model that carries biofilm-derived and autochthonous sewage taxa as a separate compartment [5], which would test whether the residual site-to-site variance left by tobamovirus normalization is biological or infrastructural.

References

[1] Lipovac, J., Sikic, M., Vicedomini, R., Krizanovic, K.. "MADRe: Strain-Level Metagenomic Classification Through Assembly-Driven Database Reduction". bioRxiv. 2025-11-10T00:00:00.000Z. URL: https://doi.org/10.1101/2025.05.12.653324

[2] Galaras, A., Chasapi, I. N., Aplakidou, E., Chasapi, M. N., Lamari, E., Diplari, S., Georgakopoulos-Soares, I., Karatzas, E., Baltoumas, F. A., Kyrpides, N., Pavlopoulos, G.. "WasteFams: A database of protein families from global wastewater microbiomes". bioRxiv. 2026-05-12T00:00:00.000Z. URL: https://doi.org/10.64898/2026.05.08.723720

[3] Belmann, P., Osterholz, B., Kleinboelting, N., Puehler, A., Schlueter, A., Sczyrba, A.. "Metagenomics-Toolkit: The Flexible and Efficient Cloud-Based Metagenomics Workflow featuring Machine Learning-Enabled Resource Allocation". bioRxiv. 2024-11-12T00:00:00.000Z. URL: https://doi.org/10.1101/2024.10.22.619569

[4] Brown, C. L., Rumi, M. A., McDaniel, L., Maile-Moskowitz, A., Sein, J., Nguyen, L., Choi, M., Hindi, F., Mullet, J., Emon, M., Moumi, N. A., Blair, M. F., Davis, B. C., Rao, J., Baffoe-Bonnie, A., Vikesland, P., Pruden, A., Zhang, L.. "Metagenomics disentangles epidemiological and microbial ecological associations between community antibiotic use and antibiotic resistance indicators measured in sewage". medRxiv. 2024-12-12T00:00:00.000Z. URL: https://doi.org/10.1101/2024.12.11.24318846

[5] Darling, A., Sastry, S., Bowie, K., Luhung, I., Franklin, A., Morley, V., Stephenson, N., Katz, D., Gratalo, D., Simas, A., Burke, T., Ruedaflores, M., Roberts, S., Turner, P., Martinello, R., Peccia, J., Healy, H. G.. "Biofilm Contributions of Bacterial Pathogens and Antimicrobial Resistance Genes to Wastewater Surveillance Signal at the Hospital Scale". medRxiv. 2026-06-29T00:00:00.000Z. URL: https://doi.org/10.64898/2026.06.24.26356348

[6] Weng, Y., Moyne, O., Walker, C., Haddad, E., Lieng, C., Chin, L., Rahman, G., McDonald, D., Knight, R., Zengler, K.. "A Multi-Omics Processing Pipeline (MOPP) for Extracting Taxonomic and Functional Insights from Metaribosome Profiling (metaRibo-Seq) data". bioRxiv. 2026-03-14T00:00:00.000Z. URL: https://doi.org/10.64898/2026.03.11.710980

[7] Ruth, N., Domman, D., Youtsey, B., Li, P.-E., Hatch, A. J., Chain, P., Shakya, M.. "Global untreated wastewater hosts a vast reservoir of previously uncharacterized microbial lineages". bioRxiv. 2026-07-30T00:00:00.000Z. URL: https://doi.org/10.64898/2026.07.28.741345

[8] Gangwar, P., Katte, P., Bhatt, M., Turakhia, Y.. "WEPP: Phylogenetic Placement Achieves Near-Haplotype Resolution in Wastewater-Based Epidemiology". medRxiv. 2025-06-18T00:00:00.000Z. URL: https://doi.org/10.1101/2025.06.09.25329287

[9] Langenfeld, K., Hegarty, B., Vidaurri, S., Crossette, E., Duhaime, M. B., Wigginton, K.. "Development of a quantitative metagenomic approach to establish quantitative limits and its application to viruses". bioRxiv. 2024-11-08T00:00:00.000Z. URL: https://doi.org/10.1101/2022.07.08.499345

[10] Ali, J., Bellankimath, A. B., Opgard, S. T., Manivannan, E. V., Simonsen, G. S., Ahmad, R.. "From culture to clarity in four hours: accelerating clinical management of bloodstream infections using metagenomics". bioRxiv. 2026-07-20T00:00:00.000Z. URL: https://doi.org/10.64898/2026.07.20.739511

[11] Bian, K., Busch, A., Norton, J., Bott, C., Gonzalez, R., Curtis, K., Tolofari, D., Khunjar, W., Graham, K. E., Pinto, A. J.. "Quantitative metagenomics using a portable protocol.". bioRxiv. 2025-10-28T00:00:00.000Z. URL: https://doi.org/10.1101/2025.10.28.684933

[12] Lataretu, M., Krautwurst, S., Huska, M. R., Marquet, M., Viehweger, A., Braun, S. D., Brandt, C., Hölzer, M.. "Targeted decontamination of sequencing data with CLEAN". bioRxiv. 2025-01-24T00:00:00.000Z. URL: https://doi.org/10.1101/2023.08.05.552089

[13] Zentgraf, J., Schmitz, J. E., Rahmann, S.. "Cleanifier: Contamination removal from microbial sequences using spaced seeds of a human pangenome index". bioRxiv. 2025-07-14T00:00:00.000Z. URL: https://doi.org/10.1101/2025.06.24.661305

[14] Xiao, A., Besse, K., Connors, D., Vian, T., Stylinski, J., Mannion, A., Lacirignola, J.. "Metagenomic Sequencing for Wastewater-Based Surveillance: Modeling and Experimental Approaches for Determining Limit of Detection". medRxiv. 2026-08-21T00:00:00.000Z. URL: https://doi.org/10.64898/2026.08.18.26360688

[15] Dylan Clark-Boucher, Brent A Coull, Harrison T Reeder, Fenglei Wang, Qi Sun, Jacqueline R Starr, Kyu Ha Lee. "Group-wise normalization in differential abundance analysis of microbiome samples". arxiv. 2024-11-26T00:00:00.000Z. URL: https://doi.org/2411.154

[16] Justen, L. J., Zulli, A., Kantor, R. S., Linfield, R. Y., Moskatel, L. S., Cunningham-Bryant, D., Kaufman, J., Johnson, M. C., McLaren, M. R., Sabeti, P.. "Establishing wastewater metagenomics as a quantitative pathogen monitoring tool with normalization". medRxiv. 2026-07-15T00:00:00.000Z. URL: https://doi.org/10.64898/2026.07.14.26356442

[17] Oskolkov, N.. "Refining filtering criteria for accurate taxonomic classification of ancient metagenomic data". bioRxiv. 2025-04-05T00:00:00.000Z. URL: https://doi.org/10.1101/2025.03.31.646431

[18] Sun, S., Fodor, A. A.. "Correction for spurious taxonomic assignments of k-mer classifiers in low microbial biomass samples using shuffled sequences". bioRxiv. 2025-06-24T00:00:00.000Z. URL: https://doi.org/10.1101/2025.06.18.660363

[19] Kim, S., Garcia, D., McCormack, C., Tham, R. X., O'Brien, M., Fuhrmeister, E., Nelson, K. L., Kantor, R., Pickering, A.. "Solid evidence and liquid gold: trade-offs of processing settled solids, whole influent, or centrifuged influent for co-detecting viral, bacterial, and eukaryotic pathogens in wastewater". medRxiv. 2025-03-05T00:00:00.000Z. URL: https://doi.org/10.1101/2025.03.01.25323156

[20] Wolfe, M. K., North, D., Jaffe, A., Zulli, A., Duong, D., Hughes, B., Goldman, M., Richardson, M., Thana, P., Chan-Herur, V., Kheradpour, P., Bidwell, A. L., Hilton, S. P., Conforti, S., Paulos, A. P., Boehm, A.. "Prospective metagenomic sequencing of wastewater across the United States yields robust viral enrichment and concordance with digital PCR measurements". medRxiv. 2026-05-08T00:00:00.000Z. URL: https://doi.org/10.64898/2026.05.07.26352651

[21] Parker, M. C., Apinda, N., Uamngamsup, S., Phetra, P., Yano, T., Sangkakam, K., Eiamsam-ang, T., Muenthaisong, A., Phetra, P., Huang, A. T., Smolinski, M., Olson, S. H., Grubaugh, N. D., Phelps, M., Rosario, K., Gil, J., Susumpow, P.. "Targeted metagenomic sequencing with spiked primers for enrichment of viruses in wastewater for pathogen surveillance". bioRxiv. 2025-12-26T00:00:00.000Z. URL: https://doi.org/10.64898/2025.12.26.690930

[22] Rushford, C., Gregory, D. A., Copen, E. E., Naydenov, J., Tovar-Mendez, A., Li, P.-E., Kantor, R. S., Shakya, M., Ruth, N., Chain, P. S. G., Hsieh, H.-Y., Hunter, T., Kome, J., Frank, M., Lyddon, T. D., Kaufman, J., O'Connor, D. H., Johnson, M. C.. "Untargeted longitudinal ultra deep metagenomic sequencing of wastewater provides a comprehensive readout of expected and unexpected viral pathogens". medRxiv. 2026-01-27T00:00:00.000Z. URL: https://doi.org/10.1101/2025.10.27.25338874

[23] Justen, L. J., Rushford, C., Hershey, O. S., Floyd-O'Sullivan, R., Grimm, S. L., Bradshaw, W. J., Bhasin, H., Rice, D. P., Stansifer, K., Faraguna, J. D., McLaren, M. R., Zulli, A., Tovar-Mendez, A., Copen, E., Shelton, K. K., Amirali, A., Kannoly, S., Pesantez, S., Stanciu, A., Quiroga, I. C., Silvera, L., Greenwood, N., Bongiovi, B., Walkins, A., Love, R., Lening, S., Patterson, K., Johnston, T., Hernandez, S., Benitez, A., McCarley, B. J., Engelage, S., Pillay, S., Calender, C., Herring, B., Robinson, C., Monett Wastewater Treatment Plant,, Columbia Missouri Wastewater Treatment Plant,, . "Deep untargeted wastewater metagenomic sequencing from sewersheds across the United States". medRxiv. 2026-10-19T00:00:00.000Z. URL: https://doi.org/10.64898/2026.03.05.26345726

[24] Kantor, R. S., Shakya, M., Ruth, N., Rothman, J. A., Rushford, C., Gregory, D. A., Epstein, A., Kaufman, J. T., Allen, J. E., Chain, P. S. G., O'Connor, D. H., Johnson, M. C.. "A genome-resolved view of the wastewater RNA virome". medRxiv. 2026-06-18T00:00:00.000Z. URL: https://doi.org/10.64898/2026.05.19.26353600

[25] Gauthier, J., Mohammadi, S., Kukavica-Ibrulj, I., Boyle, B., Landgraff, C., Goodridge, L., Levesque, R. C.. "Nanopore-based pathogen surveillance allows complete metagenome-assembled genome reconstruction of low-abundance enteric pathogens in wastewater samples". medRxiv. 2025-03-20T00:00:00.000Z. URL: https://doi.org/10.1101/2025.03.19.25324250

How this survey was made

This survey answers one question: “How are computational pipelines for untargeted wastewater metagenomics handling host and bacterial read depletion, taxonomic classification, and abundance normalization across large multi-site sequencing datasets?”

We keep our own collection of recent preprints and refresh it every day from bioRxiv, medRxiv, arXiv, PsyArXiv, SocArXiv and ChemRxiv. For each survey, we search that collection for recent work and keep the strongest matches: papers that use the same terms as the question, and papers on the same topic in different words.

The text was written from the titles and abstracts of those papers; the full texts were not read. Statements that draw on a particular paper are marked with a number in brackets. The reference list just above is the 25 papers whose work this text draws on, chosen from the 99 papers those searches returned; the rest of the shortlist is not printed. It is not a complete review of everything published on the question.

Treat this as a starting point and a reading list, not a peer-reviewed review. The searches behind this survey ran on October 21, 2026 and covered work from the previous two years.