Sequencing depth for microbiome studies: a tropical and SE Asia practical guide

How deep should you sequence each sample? The answer is sample-type dependent, and tropical and Southeast Asian samples typically need more depth than the canonical temperate references suggest. A practical guide with specific anchors for gut, soil, water, aquaculture, and fungal ITS work.

The asymmetric cost of getting depth wrong

Sequencing depth (how many reads you generate per sample) is one of two decisions that lock the ceiling on what a microbiome study can detect. The other is primer or region choice. Get either wrong and no downstream pipeline will recover what the experiment never captured.

The cost is asymmetric. Under-sequence and lose biological information permanently. Over-sequence and the budget would have been better spent on another sample, another condition, another timepoint. Studies with substantial depth waste are more common than studies that are clearly under-sequenced, but the under-sequenced ones are the ones that cannot be salvaged.

Decisions about depth are made before any bioinformatics begins. By the time the FASTQ files arrive at our hands, the ceiling is set. That is why we recommend a 30-minute design-review call before any cohort is committed to the sequencer. See Designing a microbiome study for the companion piece on primer choice.

How to read a rarefaction curve correctly

A rarefaction curve plots the number of unique OTUs or ASVs you observe (y-axis) against the sub-sampled depth (x-axis). Steep slope means you would observe more taxa with more reads. Plateau means the recoverable diversity at the chosen depth has been saturated.

For a study to be defensibly powered, the rarefaction curve should plateau before the sequencing depth you choose. If the curve is still rising at your final depth, rare taxa have been under-sampled.

A common pitfall: running rarefaction on already-rarefied data, which produces a misleadingly flat curve. Always rarefy from raw counts.

Per-sample-type depth anchors

Human gut samples are relatively low-complexity. 25,000 to 50,000 reads per sample typically saturates genus-level diversity; 50,000 to 100,000 reads gets robust species-level resolution via ASV-based denoising. The American Gut Project (McDonald et al., 2018) found genus-level rarefaction plateaued around 25,000 reads.

Soil samples are at the other end of the diversity spectrum, orders of magnitude more taxa, with most of them rare. Plan for 100,000 to 500,000 reads per sample if you want to recover rare taxa or run robust beta-diversity comparisons. The classic Lundberg et al. (2012) work suggested ~200,000 reads as a practical anchor for rhizosphere soil.

Marine and freshwater samples vary enormously by biomass. Coral-, gut-, or biofilm-associated samples can saturate around 50,000 to 100,000 reads. Open-water samples with low biomass typically need 100,000 to 200,000.

Fungal ITS amplicons add another wrinkle: ITS region length varies between species, and copy number per genome is highly variable. Under-sequencing distorts community composition rather than just truncating it. Plan for 30,000 to 100,000 reads, with the higher end recommended for tropical environmental work.

Other markers (COI for invertebrate metabarcoding, rbcL for plant DNA) need higher depth still: 50,000 to 150,000 reads per sample for environmental cohorts where universal primers are imperfect.

Why tropical and Southeast Asian samples may need more

Many of the early published depth recommendations were established on temperate-zone reference cohorts. Tropical microbiomes are often more diverse than temperate ones, more taxa in the community, longer tails of rare species, more cryptic strain variation, though the gap depends heavily on sample type and study question. The same depth that saturates a temperate-zone study can leave a tropical version visibly under-sampled.

Working anchors we tend to start from for Southeast Asian projects: tropical soils we generally aim 50 to 100 percent above the temperate baseline. For aquaculture ponds with active feed turnover and biofilm communities, the 100,000 to 300,000 reads-per-sample range tends to work. For Singapore wastewater treatment plant samples we have profiled, the 150,000 to 250,000 reads range has produced PERMANOVA-stable community profiles, though this varies with plant size and process. Tropical leaf-litter and forest-soil ITS communities often have long tails. We treat 100,000 reads as a working floor for these sample types.

These are starting anchors rather than published rules. They emerge from running studies and re-running rarefaction across different sample types in the region; published reference tables for these specific tropical adjustments do not yet exist. The point is simply that if you are designing a tropical cohort and only have temperate-zone anchors as a reference, it is worth re-checking with a pilot.

The pilot strategy: cheap insurance against the wrong depth

A practical insurance against getting depth wrong is a small pilot before the full cohort. Run 5 to 10 representative samples at the highest depth you can afford. Subsample those FASTQ files computationally to a range of depths, say 25k, 50k, 100k, 200k. At each depth, generate the rarefaction curve. The lowest depth at which the curve plateaus for your samples is a defensible cohort depth, with a 20 to 30 percent safety margin on top.

A pilot of 10 samples typically costs a small fraction of re-running a missed cohort. In our experience, most pilots surface at least one decision worth revisiting before the full run, depth, primer choice, sample handling, or batching strategy.

For routine work in well-characterised domains (gut microbiome at standard depths, V4 16S amplicon) pilots are sometimes redundant. For new sample types, new geographies, or unusual cohort designs, pilots are usually worth the time.

When 100k is enough vs when you need 500k

Different downstream questions have different depth requirements. Alpha-diversity comparisons (Shannon, Simpson) are relatively forgiving, once the rarefaction curve plateaus, more depth adds little. Beta-diversity ordination (PCoA, NMDS) and PERMANOVA are similarly forgiving for the dominant signal but unstable for borderline differences if rare taxa are under-sampled.

Rare-taxa detection, pathogen surveillance, biomarker discovery, functional gene tracking, needs more depth. If you care about the bottom 5 percent of the community, you cannot under-sample.

Differential abundance testing is moderately depth-sensitive. Functional inference (predicted function for 16S, pathway mapping for shotgun) is moderately depth-sensitive. Machine learning for biomarker classification is depth-hungry, model accuracy drops noticeably below 50,000 reads per sample for most cohort sizes.

Coupling depth with primer choice

These two decisions are not independent. A short region like V4 produces tightly defined ASVs that need less depth to characterise reproducibly. A longer region like V3–V4 produces more variants and benefits from deeper coverage to keep rare ASVs above the noise floor. ITS amplicons have variable lengths, so depth must compensate for uneven amplification across taxa.

A V4-only Singapore aquaculture study at 80,000 reads per sample is well-powered. The same study with V3–V4 may want 120,000 to 150,000. The same study with ITS-2 should sit at 100,000 to 150,000 minimum.

How Nova Lifetech changes the outcome

These are not look-up-the-answer decisions. They depend on cohort size, the statistical question, the budget, and the eventual destination of the data. Our team has designed and run microbiome studies across these sample classes for academic and industry partners; that experience compresses into specific guidance for a given project.

We work alongside the principal investigator or the operations lead from study design through delivery. The handoff at the end is not just data. It is the rationale for every decision, so the team can defend it in review, adjust it for the next round, or scale it without consulting us again.

If you are designing a study now, start the conversation, a short call is usually enough to resolve the open depth questions before sequencing is committed.

A note on the numbers in this article

The depth ranges given here are working anchors, not published rules. Where citations exist (McDonald 2018 for human gut; Lundberg 2012 for rhizosphere soil) we have noted them. The other figures (particularly the tropical and Southeast Asian adjustments) emerge from operational experience running studies in the region. They are starting points for a design conversation, not specifications. Your specific cohort, sample type, and statistical question will move the right answer up or down.

↗Designing a microbiome study: primer choice and the depth coupling
↗Amplicon and metagenomics services: what we deliver
↗Talk to us: start an enquiry
The asymmetric cost of getting depth wrong
How to read a rarefaction curve correctly
Per-sample-type depth anchors
Why tropical and Southeast Asian samples may need more
The pilot strategy: cheap insurance against the wrong depth
When 100k is enough vs when you need 500k
Coupling depth with primer choice
How Nova Lifetech changes the outcome
A note on the numbers in this article

Wastewater AMR surveillance: methods and regulatory readiness for SE Asia utilities

Wastewater is becoming a recognised public-health signal for antimicrobial resistance. What samples, what methods, and how do you make a deliverable that holds up to a regulator? A guide for utilities, public-health agencies, and environmental-health researchers in Southeast Asia.

NGS at work in industry: four sectors, one shared blueprint

NGS used to be a research expense. Now it is an operational tool. Across aquaculture, wastewater, human microbiome research, and agriculture, microbial sequencing is replacing best-guess practice with measured insight, protecting yield, reducing risk, and surfacing value that did not exist a decade ago.

Tell us what you are working on.

A short note about your samples and questions is enough to start. We reply with how we would approach it.