Biology · Genomics, Omics & Bioinformatics
Metagenomics, microbiome data and compositional challenges
Category A falls from one fifth to one tenth of an invented community, yet its complete count stays at twenty. The other categories grew in the supplied account. A relative decrease therefore does not require an absolute decrease, and a fixed-size sequence library cannot recover the missing total by arithmetic alone.
Gene Expression: RNA & Proteins explains how sequence, RNA and protein are related. Phylogeny, Classification & the Tree of Life introduces ancestry and the interpretation of trees. Research Design, Data & Scientific Practice explains sampling, independent units and the limits of comparisons. Their English chapters are available in this working manuscript; publication remains pending. The examples here use letter strings, inclusive coordinates, averages, fractions and ordinary graph reading. Each convention is introduced before it is used.
Back to the Biology mapBy the end, you can
Interpret community sequence records and relative composition while separating read counts, absolute abundance and evidence of biological activity.
A community sample contains more than one origin
Metagenomics studies sequence information from mixed community material. Reads may represent different members and regions, and some cannot be assigned confidently. An assembled community fragment is not automatically a complete genome of one known organism. Shared regions and incomplete records can complicate the connection between a sequence and its origin.
A targeted marker survey asks a narrower question than a broad metagenomic sequence account. It may help distinguish categories through a selected region while leaving much of their genomic content unobserved. The recorded material, reference collection and assignment method therefore belong to the interpretation of any category table.
The evidence-layer model gives one imaginary feature twenty assigned DNA reads and another zero. No RNA, protein or activity measurement is supplied. A DNA-based functional prediction can suggest potential capacity, but it does not measure expression, protein activity or the rate of a process in the community.
DNA observations leave other layers open
Read the diagram in words
Feature one has twenty DNA reads and feature two zero. Every RNA, protein and activity entry is unobserved. The zero is a supplied library count; null marks a layer without an observation.
| Assigned sequence feature | Assigned DNA read count | RNA measurement | Protein measurement | Activity or process-rate measurement |
|---|---|---|---|---|
| Feature one | 20 | Not observed in this record | Not observed in this record | Not observed in this record |
| Feature two | 0 | Not observed in this record | Not observed in this record | Not observed in this record |
Observed assigned-read count of zero
Not observed in this record
Sequence evidence alone does not measure expression or activity
Original fictional feature-read account: twenty and zero assigned DNA records, other measurement layers unobserved. No actual taxon, gene, functional annotation confidence, viability inference, detection probability or activity measurement.
A composition preserves ratios while losing scale
A relative composition divides each part by the total represented in that account. Its parts sum to one. If every absolute count is multiplied by the same factor, the composition stays unchanged. Conversely, one category's fraction can change when other categories change, even if its own absolute count does not.
The original complete accounts are twenty/thirty/fifty and twenty/sixty/one hundred twenty. Their totals are one hundred and two hundred. Category A remains at twenty, B doubles and C increases more than twofold. Their second fractions are one tenth, three tenths and six tenths. These are complete invented counts, not estimates recovered from a sequence library.
A third account doubles every count in the second. It has total four hundred and the same composition as the second. A hypothetical library with one hundred assigned category reads could therefore have the same ten/thirty/sixty composition for either account under an ideal proportional representation. That illustration does not assume that real read fractions equal cell fractions without bias.
A falling fraction can accompany an unchanged count
Read the diagram in words
Counts 20/30/50 total 100, 20/60/120 total 200, and 40/120/240 total 400. Fractions are .2/.3/.5 then .1/.3/.6 twice. Category A stays at 20 in the first two accounts while its fraction halves; the second and third fractions match despite doubled counts.
| Assigned complete community account | Anonymous category | Complete count within the stated boundary | Total across all three categories | Category count divided by that account’s total | Idealised assigned category reads in a one-hundred-read library |
|---|---|---|---|---|---|
| First complete account | Category A | 20 | 100 | 20/100 = 0.2 | 20 |
| First complete account | Category B | 30 | 100 | 30/100 = 0.3 | 30 |
| First complete account | Category C | 50 | 100 | 50/100 = 0.5 | 50 |
| Second complete account | Category A | 20 | 200 | 20/200 = 0.1 | 10 |
| Second complete account | Category B | 60 | 200 | 60/200 = 0.3 | 30 |
| Second complete account | Category C | 120 | 200 | 120/200 = 0.6 | 60 |
| Third complete account | Category A | 40 | 400 | 40/400 = 0.1 | 10 |
| Third complete account | Category B | 120 | 400 | 120/400 = 0.3 | 30 |
| Third complete account | Category C | 240 | 400 | 240/400 = 0.6 | 60 |
| Assigned complete community account | Idealised assigned category reads in a one-hundred-read library |
|---|---|
| First complete account | 100 |
| Second complete account | 100 |
| Third complete account | 100 |
| Complete count within the stated boundary | 0 — 400 |
|---|---|
| Category count divided by that account’s total | 0 — 1 |
A library composition does not recover the absolute total
Complete invented category counts within the same defined boundary; all quantitative totals are supplied independently of sequencing. Library counts are ideal assigned proportional illustrations, not estimates of real cell fractions. No sampling/handling protocol, bias-free sequencing assumption or biological interaction.
Identify what additional information is missing
Read totals describe recorded sequence observations. Absolute organism abundance instead needs a defined boundary or amount of sampled material and appropriate quantitative evidence. Genome sizes, marker copy numbers, extraction and assignment biases can affect the relationship between cell abundance and observed read fractions. Increasing read depth alone does not supply all this information.
A zero assigned read count is an observation about the library and assignment rule. It is not necessarily proof that the organism or feature is absent from the community. Detection can be limited by sampling, sequence coverage or missing reference information. The model does not provide a detection probability or a reason for its zero.
Comparisons should retain the denominator, the category definition and the measurement layer. A relative association may generate a biological question without establishing an interaction, a disease mechanism or a direction of absolute change. Combining composition with appropriately scoped quantitative evidence can address a richer question; the composition alone should not be credited with that evidence.
Worked example
Compare the parts and the whole separately
The first complete total is twenty plus thirty plus fifty, or one hundred. The second is twenty plus sixty plus one hundred twenty, or two hundred. A's fraction changes from 20/100 to 20/200: one fifth to one tenth, even though its count remains twenty.
B remains at fraction three tenths while its count increases from thirty to sixty. In the third account the counts are forty, one hundred twenty and two hundred forty. Divide each by four hundred to obtain the same fractions as in the second account.
These examples show that neither a changing nor an unchanged fraction uniquely specifies an absolute change. The separate DNA record adds no RNA or activity data. Its zero and null entries must remain different: zero is the supplied read count, while null marks a layer not observed.
Check your understanding
Did category A decline in absolute count between the first and second complete accounts?
Show reasoning
No. It remains at twenty. Its fraction falls because the complete total rises from one hundred to two hundred. A statement about the fraction must not be silently rewritten as an absolute decline.
Can the identical second and third compositions identify which total was present?
Show reasoning
No. Their counts differ by a common factor of two, which cancels in each proportion. Additional quantitative information is needed to distinguish the absolute totals.
Do twenty DNA reads establish current expression or activity of that feature?
Show reasoning
No. The record contains DNA evidence only. Expression, protein state and activity are different measurements; the model explicitly leaves those layers unobserved.
What to carry forward
Keep composition, absolute abundance, detection and activity separate; each needs its own evidence and denominator.
References
Resources supporting this lesson’s explanations and examples.
- Metagenomics (Official glossary, inspected 13 October 2026)NHGRI; Julie Segre narrationAgency glossary
- Microbiome Datasets Are Compositional: And This Is Not Optional (Frontiers in Microbiology 8:2224 (2017). DOI: 10.3389/fmicb.2017.02224)Gregory B. Gloor, Jean M. Macklaim, Vera Pawlowsky-Glahn and Juan J. EgozcueOriginal mini-review
- Quantitative microbiome profiling links gut community variation to microbial load (Nature 551(7681):507–511 (2017). DOI: 10.1038/nature24460)Doris Vandeputte and colleaguesResearch abstract