Soil microbial indicator species analysis looks for taxa or sequence features that are consistently associated with particular habitats, treatments or environmental groups. If you are reading an IndVal score, the indicspecies multipatt output, specificity A, fidelity B, permutation p values or a list of candidate indicator ASVs, begin by asking how the site groups were defined and which microbial features were measured.
An indicator taxon is not automatically beneficial, harmful, unique to an ecosystem or capable of diagnosing a soil condition with certainty. Indicator value combines information about how strongly a feature is associated with a group and how consistently it appears among that group’s sampled sites. The observed association is conditional on the sequencing method, comparison groups, sampling design and analysis rules. It is evidence about the study’s data, not a universal taxonomic law.
This guide explains soil microbiome indicator species analysis from the original feature table through IndVal, group-equalised IndVal.g, multiple-site-group combinations, significance testing, adjusted p values, ecological interpretation and real-world communication. The examples are invented for teaching, with arithmetic shown explicitly rather than presented as measurements from an actual soil experiment. The aim is to preserve what a meaningful association can tell us while avoiding claims the statistic does not justify.
Quick answer: what makes a good indicator in a study?
A candidate microbial indicator is a feature whose measured pattern associates with one site group or a defined combination of groups. An indicator with strong specificity is disproportionately represented in the target group compared with other sampled groups. An indicator with strong fidelity occurs consistently across sites in that group. The combination can highlight taxa useful for describing environmental differences and generating biological hypotheses.
The term ‘indicator’ here describes statistical association under defined sampling and feature representations. It is not a laboratory diagnosis and need not imply a causal relationship. An organism can be a strong indicator because it thrives in a condition, tolerates it, is correlated with another environmental factor or is preferentially detected under the study’s preparation conditions. Every interpretation begins with the original research design.
Define the habitat or treatment groups first
Imagine a study comparing well-drained soil, waterlogged soil and a periodically flooded transition zone. Indicator species analysis requires a clear group label for each independent biological sample. The group definition determines what association is being investigated. Grouping by drainage regime answers a different question from grouping by soil pH, land use or plant species, even if the same samples are analysed.
Ask whether groups were specified before seeing the microbiome table or created from patterns in the very data being tested. Circular definitions can make apparent indicator strength less convincing. Also check whether the field design separates drainage from confounded differences such as season or geographic location. If each group occupies one site, a taxon may indicate site identity rather than a transferable environmental mechanism.
The sampling unit controls what is independent
A soil core, a composite of several subsamples, and a pooled DNA extract are not interchangeable biological replicates. Six sequencing libraries prepared from one original soil sample do not represent six independent sites. A feature appearing in each technical repeat demonstrates technical consistency under the workflow, not broad ecological fidelity across the landscape.
Inspect whether samples came from independent plots, repeated visits to the same plot or nested depths within sites. If the same plot is sampled monthly, analyses should recognise this dependence. A permutation test that freely shuffles all repeated samples can generate misleading evidence. Statistical association must respect how biological units were collected and assigned to groups.
From organism to feature table: the measurement chain
An indicator analysis needs a community table containing observations of candidate taxa or features across samples. In an amplicon project, the rows or columns may represent inferred ASVs, not proven species. Extraction, primer choice, sequencing depth, denoising, filtering and classification determine which features appear. The label ‘indicator species’ is a historical method name; it does not guarantee every table feature is taxonomically resolved to species.
Read the soil amplicon ASV guide when feature identity is unclear. A result for ASV-42 can be meaningful even if its classifier supports only family-level taxonomy. Report ‘indicator ASV’ rather than inventing a species name. Conversely, collapsing several ASVs to one genus can conceal important patterns, so the aggregation level should match the question.
Presence is not the same as abundance
A presence–absence matrix records whether a feature was detected in each sample under the pipeline’s detection rules. An abundance table records counts, relative profiles or another numeric measure. These representations contain different information. A feature observed at one read in every target sample may have high observed fidelity while contributing few reads. Another feature may be extremely abundant in two target samples but absent from the others.
Indicator metrics often combine a specificity component with a fidelity component precisely because these dimensions can differ. Do not describe a high abundance in one soil core as evidence that an organism reliably indicates the entire habitat. Likewise, repeated detection at very low counts requires careful attention to contamination, read depth and detection limits before it is presented as an ecological signature.
What the IndVal method is designed to capture
The classic Dufrêne–Legendre indicator value combines measures of how characteristic a feature is of a group and how faithfully it occurs across that group’s sampled sites. Under common conventions, an A component represents specificity and a B component represents fidelity. Their product produces an association value under a defined scaling; some software displays or tests a transformed form such as a square root.
An IndVal result can be useful for identifying candidate taxa associated with vegetation types, soil moisture regimes or restoration stages. It does not explain why the association exists. The index describes a pattern among measured feature values and predefined groups. Explanation requires additional environmental data, careful causal design or laboratory experiments.
Specificity A asks how concentrated the association is
In an abundance-based interpretation, specificity compares a feature’s group-associated mean abundance with its corresponding means in other groups under the chosen IndVal variant. If a feature has a mean abundance of 18 units in target group A and six in group B, a simple two-group specificity ratio is 18 divided by 24, giving 0.75. This means three quarters of that feature’s group-mean total lies with A on that scale.
The value is not ‘75% of all bacteria are this taxon.’ It describes the relative allocation of the feature across the compared groups. The precise calculation can vary with group weighting and software conventions, particularly between IndVal and IndVal.g. Ask whether the input is counts, relative abundance or another transformation. The resulting A component inherits those choices.
Fidelity B asks how consistently a feature is detected
Fidelity describes the proportion of sampled sites in the target group where the feature meets the presence criterion. If the feature is detected in six of eight independent group-A plots, B is six divided by eight, or 0.75. If the feature occurs in every target plot, B can reach one, but that observation is restricted to the sampled plots and detection method.
A B of 0.75 does not guarantee future detection in three of every four soils. The observed sample may be small, clustered, collected in one season or affected by laboratory sensitivity. Wider prediction needs independent validation across locations and dates. The mathematical meaning is sample-level consistency under the observed design, not a probability guaranteed by nature.
Worked example: two components and a combined score
Combine the illustrative values A equals 0.75 and B equals 0.75. Their product is 0.5625. If a particular software output reports the square root of the product, it would display 0.75 rather than 0.5625. Both numbers can refer to the same underlying association under different reporting conventions. Read the function documentation and scale before comparing results.
The calculation is pedagogical, not a universal formula for every IndVal variant. Group-equalised calculations and allowed group combinations can change the underlying A term and the best matching association. The central idea remains: a useful indicator should characterise the target group and occur consistently within it. A strong value is not a measure of soil function or conservation importance.
A high A and low B tell a different story
Consider a feature that is highly concentrated in a wetland group when present but appears in only one of ten wetland plots. Its specificity may be high, but its fidelity is low. The feature may be associated with a very narrow wetland microhabitat, a seasonal episode, a sequencing artefact or simply a rare organism. Calling it a reliable wetland-wide indicator would exaggerate the repeatability demonstrated.
This pattern might still be scientifically interesting. It may motivate a targeted study of microhabitats or transient disturbance. The correct description is ‘strong group concentration in the observed detections, but poor within-group occurrence.’ Separating components allows readers to see whether the association is widespread and consistent or intense but sporadic.
A high B and low A are also informative
Now imagine a feature detected in every wetland plot but also common in grassland and forest soils. Its target-group fidelity is perfect, yet its specificity is weak. It may be a widespread generalist rather than a marker that distinguishes wetland soils. The observation can still tell us something about ecological tolerance, but it has limited power for identifying one habitat relative to others.
This is why simply counting positive samples is insufficient. A consistently detected taxon can be uninformative as a discriminating indicator if it occurs everywhere in the comparison. Report the compared groups explicitly. An indicator’s specificity is relative to the alternatives included in the study and may change when the reference set of environments changes.
The meaning of IndVal.g and group equalisation
Software packages such as indicspecies implement several related association measures. IndVal.g applies a group-equalised form of specificity, helping address differences in the number of sites assigned to groups under the defined method. The distinction matters when comparing, for example, twenty dry-plot samples with only five wet-plot samples. The plain IndVal and IndVal.g outputs need not match.
Do not assume the suffix ‘.g’ is decorative. Read the actual method used, group sizes and formula definition before giving a numerical explanation. In the indicspecies documentation, different quantities and square-root transformations may appear in stored fields. Report the software, function and variant, not merely ‘an indicator score of 0.8’ with no context.
How multipatt can consider group combinations
The indicspecies multipatt function can study association with combinations of site groups rather than only one group at a time. A microbial feature may characterise both well-drained and intermittently drained soils but not permanently flooded soils. Such a feature could be associated with the union of the first two groups under the selected model. That is a different research question from identifying a taxon unique to one habitat.
The allowed combinations may be restricted for ecological reasons. A researcher should not assume every mathematically possible union makes biological sense. A result indicating association with two groups does not imply the feature is equally abundant across them or absent elsewhere without examining the component table. Preserve the group-combination label and explain what it actually covers.
One group versus several groups changes the reader’s job
An indicator tied to a single predefined treatment may help describe a specific management-associated community pattern. An indicator tied to a combination of treatments may reveal a common environmental characteristic shared by those conditions. These are both legitimate outcomes, but they answer different questions. A study should clarify whether it sought single-group diagnostics or broader habitat associations.
If an analysis unexpectedly identifies a combined group, the next step is to examine measured environmental variables rather than invent a unifying ecological mechanism. A shared feature might respond to pH, plant cover, soil depth, moisture or another factor correlated across the included groups. The result is a clue for a better experiment, not automatically an explanation of the shared biology.
What a permutation p value means
A permutation procedure compares an observed group-association statistic against results obtained under rearrangements of group labels or observations consistent with the null hypothesis and sampling design. A small p value indicates the observed association would be unusual under the chosen null model and allowable rearrangements. It does not prove causation, diagnose soil health or guarantee that an indicator will work in a new region.
The validity of the permutation depends on exchangeability and independence. Repeated measurements, blocked field experiments and nested sampling can restrict how samples should be shuffled. A p value computed from an invalid permutation design may sound precise while reflecting a flawed comparison. Ask how permutations respected the field layout before interpreting statistical significance.
The reported p value may not be adjusted
The indicspecies multipatt documentation notes that p values in its standard output are not corrected for multiple testing. A microbiome study may examine hundreds or thousands of ASVs, creating many chances for apparently small p values. If the analysis reports only the most striking unadjusted associations, its false-positive burden can be substantial.
Check whether an appropriate false-discovery-rate adjustment or other multiple-testing strategy was applied across the intended hypothesis family. Statistical adjustment does not correct confounding, poor sampling or laboratory artefacts, but it addresses part of the multiplicity problem. A transparent article identifies the number of features tested and avoids presenting nominal significance as a clean guarantee of biological reliability.
Why a strong score can be nonsignificant
An indicator might have a high observed association score but occur in very few independent samples. With limited replication, the permutation distribution may not provide compelling evidence that the pattern generalises beyond chance under the null model. Conversely, a moderate but consistent association in a large design can be statistically detectable. Association strength and statistical evidence are related but different descriptions.
Report effect or association strength alongside p values, uncertainty and group sizes. Do not select indicators by significance alone, and do not select them by the highest score alone. Ask whether the observed pattern is biologically interpretable, technically robust and stable across reasonable analysis choices. A reliable teaching example shows all these dimensions rather than ranking taxa by one impressive number.
What does a zero in an ASV table mean?
A zero generally means the pipeline retained no reads assigned to that feature in a given library. It does not prove the organism was absent from the original soil. Limited sequencing depth, extraction losses, PCR primer mismatches and filtering can produce nondetection. An indicator’s fidelity may therefore depend on detection sensitivity, particularly for rare taxa.
When an indicator appears sporadically, inspect read depth and quality control before labelling it ecologically rare. A sample with few retained reads is less likely to detect uncommon features. Some analyses deliberately filter low-prevalence ASVs, changing the candidate indicator set. The presence criterion and filtering policy belong in the methods, because they define the B component and its uncertainty.
Compositional data require careful language
Amplicon count tables are shaped by sequencing effort and amplification; relative-abundance profiles are closed to a total of 100%. A feature’s share can rise because other features fall, even without an increase in its absolute gene copies per gram of soil. Indicator analysis applied to these tables inherits the compositional limitations of the measurement. Specificity based on relative values is not an absolute count of living cells.
Write ‘indicator ASV associated with these analysed libraries’ rather than ‘indicator organism grew by this percentage’ unless an independent quantitative bridge supports it. A strong association with a soil habitat can be real while its abundance scale remains relative. Keeping the denominator visible prevents erroneous claims about ecological population growth or decline.
Indicator analysis and differential abundance differ
Differential-abundance analyses typically investigate whether a feature’s estimated abundance differs between groups under an explicit statistical model, often reporting effect sizes such as log fold changes and adjusted p values. Indicator analysis emphasises association with one or more site groups through specificity and fidelity. These questions overlap in data source but are not identical.
A feature with a large modelled fold change may occur in only a few treatment plots and therefore have low fidelity. Another feature may be a reliable indicator across the group even when its mean difference is moderate. Use the soil differential-abundance guide to understand what that model estimates. Avoid publishing two articles claiming to answer the same question when their actual statistics serve different purposes.
Indicator analysis and beta diversity differ too
Beta-diversity ordination asks how entire community profiles differ among samples. Indicator analysis asks which particular features are associated with site-group membership. A visible separation in a PCoA plot does not by itself identify a reliable indicator feature. Similarly, one strongly associated ASV need not drive a broad multivariate separation when other features vary differently.
The beta-diversity guide explains Bray–Curtis, NMDS and PERMANOVA. These results can complement indicator analysis: the ordination describes community structure, and IndVal highlights particular candidate markers within the same sampling frame. Keep their separate statistics and assumptions visible in any synthesis.
Indicator taxa are not necessarily keystone taxa
A keystone organism is usually discussed in terms of disproportionately important ecological effects relative to its abundance or other measures. Statistical association with a soil group does not establish that role. An indicator may be easy to recognise without causing the habitat to remain stable, while a functionally important organism may be too widespread or too rare to score highly as a group indicator.
Do not call a taxon ‘keystone’ merely because it ranks first in an IndVal table. Functional interventions, interaction studies or other evidence are needed to support causal ecological roles. The distinction parallels microbial network interpretation, where a highly connected statistical node is not automatically a keystone species. A clear report differentiates association, predictive usefulness and demonstrated mechanism.
A concrete example from contrasting soil moisture
Imagine two groups, each containing eight independent plot samples: a wet group and a dry group. Candidate ASV-17 is detected in seven wet plots and one dry plot. Its mean measured abundance is much higher in wet than dry plots. These observations are compatible with strong specificity and fidelity for the wet group, provided the assay and grouping are trustworthy.
That does not prove ASV-17 prefers water physically or causes wet conditions. It might reflect differences in plant roots, oxygen availability or geographic setting that coincide with moisture. A careful summary calls it ‘a candidate wet-group indicator among the sampled sites.’ To investigate the mechanism, measure additional variables or test the organism under controlled conditions. The association points to research; it does not replace research.
A contrasting example from soil pH
Suppose another ASV appears consistently across acidic and neutral soils but rarely in alkaline soils. A multipatt analysis might associate it with a combination of the first two groups. Calling the ASV ‘an acid indicator’ would erase its neutral-soil association. A better description is that it characterises the acidic-and-neutral set relative to the alkaline set in the sampled comparison.
The example shows why compound group labels need careful translation. Terms such as ‘preferred habitat’ can overstate the result unless an experimental preference test exists. We know the statistical profile of sequence features across sampled environments; the mechanism may still be unknown. Group combinations should be described in full, including which sites are inside and outside the matched set.
Unequal group sizes and spatial clustering
A field study with twenty dry plots and four wet plots can yield an apparently stable pattern in the large group but a highly uncertain fidelity estimate in the small group. The appropriate IndVal variant and permutation structure must reflect the design. Spatially clustered sites may also share unmeasured conditions, reducing effective independent information despite a large number of core samples.
A good report lists independent biological sample counts for every group, not merely total sequencing libraries. It explains whether measurements were nested or paired and whether analyses were sensitive to a few influential samples. Without that information, a high score may be a property of the collected convenience sample rather than a transferable biological marker.
Seasonality can turn an indicator into a snapshot
Soil microbial communities change as temperature, moisture, root growth and substrate availability change. An ASV strongly associated with wet soil during one season may be common under different conditions later. Sampling all groups on one date reduces some timing differences but cannot establish year-round stability. Conversely, groups sampled on different dates are confounded with season.
If the intended application is monitoring over time, independent seasonal validation is essential. A feature should retain useful discrimination in new samples collected outside the development dataset. Report whether the indicator was tested longitudinally and how often it was detected at the intended threshold. A stable-looking score from one campaign is valuable evidence about that campaign, not a permanent ecological contract.
External validation asks whether the marker travels
An indicator derived from one set of fields may fail in another region because soil chemistry, plant communities, dispersal history and reference organisms differ. The simplest validation is to hold out genuinely independent sites, apply the predefined detection rule and evaluate performance. If the original score required reselecting thresholds or taxonomic groups for each validation set, its portability is less convincing.
Distinguish discovering indicators from using them operationally. Discovery tolerates exploratory comparisons and hypothesis formation. Practical environmental screening demands reproducibility, sensitivity, specificity and calibration for the intended decision. A high IndVal is not itself a validated diagnostic accuracy estimate, and no reader should infer that a specific threshold provides universal prediction.
Predictive sensitivity and specificity are separate quantities
The words specificity and fidelity in indicator value should not be silently equated with diagnostic specificity and sensitivity as defined from a confusion matrix. When constructing a classifier, sensitivity might be the fraction of truly target-condition sites correctly identified, and diagnostic specificity the fraction of nontarget sites correctly rejected. These require a declared classification rule and reference truth labels.
A numerical indicator score alone does not supply those operating characteristics. If a management agency wants to use an ASV as a decision marker, it should specify the detection threshold and evaluate false positives and false negatives in independent data. Statistical association is an important first step, not a completed field-test protocol.
The taxonomic naming layer can mislead
A result table might label an indicator feature only as ‘unclassified family’ or assign a broad genus with limited confidence. Some readers assume a genus assignment applies uniformly to every member’s biology. That can be wrong. Marker sequences may not resolve species, and strains can vary in gene content and ecological function. Report the supported rank and keep the original ASV identifier.
For educational prose, explain both the readable label and the underlying evidence. The taxonomic name helps orientation, but the sequence feature allows the association to be traced across tables or future reclassifications. If a reference database changes the name, the indicator analysis can still be tied to the original feature rather than rewritten as a new organism discovery.
Source contamination can create a false habitat marker
Suppose wet samples were extracted using one reagent lot and dry samples another. A contaminant associated with one lot might become an apparent indicator of wet soil. A permutation test would not necessarily recognise the laboratory cause if processing batch perfectly overlaps the biological group. The feature could have strong specificity and fidelity for the wrong reason.
Review extraction blanks, PCR negatives, preparation order and sample randomisation. A laboratory artefact may be highly reproducible within a batch. Strong statistical association therefore cannot substitute for controls. If group and laboratory batch are confounded, acknowledge that the biological source of the indicator is unresolved until new balanced samples can be analysed.
Multiple candidates may reveal one underlying gradient
Several ASVs can indicate the same group because they respond to a shared environmental factor or because their relative abundances are statistically connected. Reporting ten indicator taxa does not automatically establish ten independent ecological mechanisms. Some may belong to related lineages or covary through broad community composition changes.
Examine environmental data and correlation structure. The soil microbial network guide explains why co-occurrence edges are not proof of direct interaction. IndVal addresses group association, while network approaches examine selected inter-feature statistical relationships. They are complementary descriptions, not substitutes for mechanistic experiments.
Interpreting a results table without guesswork
Begin by locating the row identifier, target group or group combination, association score, A and B components if provided, permutation p value, adjustment method and sample prevalence. Check whether the displayed value is the raw IndVal product or a square-root form. Record group sizes and feature-filtering rules. Then compare the apparent signal with individual sample values rather than trusting a ranked table alone.
A short explanation can preserve all essential context: ‘ASV-17 was associated with the wet-soil group under IndVal.g, with high observed group specificity and occurrence across the sampled plots; statistical evidence was evaluated using a design-appropriate permutation test.’ Add the actual values only after verifying the software’s definition. Avoid calling the result proof of a waterlogging mechanism.
A worked reporting repair
Consider the proposed sentence, ‘Species X is the best biological indicator of healthy soil worldwide because its p value was 0.001.’ The data actually come from twelve local plots, an ASV labelled to genus level, and several hundred unadjusted permutation tests. The sentence overstates taxonomic identity, generality, biological value and statistical certainty.
A defensible revision is: ‘In this sampled comparison, the marker feature assigned to genus X showed a strong association with the defined treatment group. Its usefulness as a transferable soil-health indicator requires independent site validation and functional evidence.’ The revised statement gives the reader a checkable pattern and the next relevant question, instead of promising global diagnostic certainty.
How to design a useful validation exercise
Select a small number of leading candidate indicators using a discovery dataset. Write down the exact feature identifiers, target group definitions and detection or abundance rules. Then apply those rules to genuinely independent soil plots from a later date or a different field, without searching again for better thresholds. Record false detections, missed target sites and unexpected changes in laboratory recovery.
Analyse whether the pattern remains stable when read depth, extraction procedure and soil context are controlled as far as practical. Report confidence intervals or other uncertainty appropriate to the sample size. If performance falls, investigate why rather than hiding the result. A failed external test may teach more about ecological context than a successful model fit to the original samples.
Translating ‘specificity’ and ‘fidelity’ for ordinary readers
In this method, specificity refers to concentration of the feature’s measured association in the target site group. Fidelity refers to consistency of detection across group members. Those technical definitions do not always match common-language meanings such as personal loyalty or exact diagnostic specificity. A plain-English explanation should define them before using the scores.
The cross-linguistic vocabulary guide helps readers see why apparently familiar words can mislead across technical contexts. The translation and meaning guide explains why preserving conceptual relationships is more important than finding quick word substitutions. Good scientific communication protects the distinction between a statistical association and an environmental diagnosis.
Frequently asked questions
Is an indicator ASV necessarily a unique species? No; marker resolution may be broader or narrower. Does high IndVal prove the taxon causes the condition? No. Can an indicator be associated with several habitat groups? Yes, when the model permits group combinations. Does IndVal.g use the same group weighting as every IndVal variant? No; read the specific method definition.
Are multipatt p values automatically multiple-testing adjusted? The documented standard output does not automatically provide that correction. Can an indicator fail on another continent or season? Yes. Is high specificity enough if fidelity is low? It may indicate an intense but inconsistent association, not a reliable general marker. Can a statistical indicator be ecologically important? Certainly, but importance requires evidence beyond the association score.
Independent practice and answers
Feature A is detected in six of eight target plots; what is observed fidelity? Six divided by eight is 0.75 under a simple presence definition. Its mean abundance is 18 in the target group and six in the only comparison group; under the simplified two-group specificity ratio A is 0.75. Their product is 0.5625. If software returns the square root, the displayed value would be 0.75. Explain which score the report uses.
Now imagine a feature present in all target and all control plots at similar levels. Is it a good single-group indicator? No, because specificity will be weak even with high target-group fidelity. Finally, consider an indicator found only in wet plots collected in winter and dry plots collected in summer. Can you claim it indicates moisture rather than season? No; the sampling design confounds those variables.
Authoritative sources and related learning
The indicspecies multipatt documentation explains supported IndVal and IndVal.g variants, site-group combinations and permutation-based association tests. The R indicspecies association documentation defines components and reference methods. Read these when a paper’s display convention is unclear; small differences in output scale matter.
Continue with eduKateSG’s guides to soil ASV tables, soil differential abundance and soil beta diversity. They answer different questions about the same underlying feature data and should be chosen according to the reader’s purpose.
The lasting lesson: association is a starting point
Indicator species analysis can identify meaningful candidate microbial features associated with a defined set of soil conditions. Its strength comes from combining how characteristic a feature is of the target group with how consistently it appears there. Its limits arise from measurement quality, study design, reference taxonomy, group definitions and the difference between statistical association and causal ecological function.
Read the A and B components, verify the variant of IndVal, inspect sample counts and multiple testing, then evaluate whether the indicator travels to independent environments. That sequence makes a ranked results table understandable without pretending that one high score diagnoses an entire soil ecosystem. A useful indicator is a carefully supported hypothesis with an explicit intended application, not an automatic label of biological goodness.