Amplicon Sequence Variant (ASV)

ASVs are defined by exact sequence identity rather than a similarity threshold. Unlike OTUs clustered at 97% similarity, an ASV distinguishes sequences differing by as little as one nucleotide. This resolution is achieved through denoising algorithms that model the error profile of the sequencing run, separating biological variation from PCR and sequencing artifacts. The result is a set of unique, biologically meaningful sequences that can resolve ecotypes and strains that would be collapsed into a single OTU.

ASV inference relies on a parametric error model that estimates base-specific substitution rates from the sequencing data itself. The DADA2 algorithm, for example, alternates between estimating error rates and partitioning sequences until convergence. This process distinguishes rare true variants from abundant erroneous sequences by evaluating the abundance p-value—the probability that a sequence is too abundant to be explained by error alone. This statistical rigor eliminates the need for arbitrary abundance filtering thresholds.

Because ASVs are defined by exact sequence strings, they are intrinsically reproducible across different datasets, sequencing runs, and laboratories. An ASV representing E. coli in one study is identical to the same ASV in another, enabling direct meta-analysis without the need to re-cluster or harmonize OTU definitions. This property makes ASVs ideal for building cumulative reference databases and for longitudinal studies where consistent tracking of specific sequence variants is critical.

ASV identification is a reference-free, de novo process. The denoising algorithm infers true biological sequences directly from the sample without aligning against an external reference database. This independence is crucial for discovering novel diversity, as ASVs can represent previously uncharacterized organisms or variants. Taxonomic assignment occurs as a separate, downstream step using classifiers like the RDP Naive Bayesian Classifier or BLAST against curated databases such as SILVA or Greengenes.

ASV tables preserve the compositional nature of sequencing data with higher fidelity than OTU tables. By resolving exact sequences, ASVs prevent the artificial inflation of diversity estimates caused by OTU clustering artifacts and reduce spurious merging of distinct lineages. This precision is essential for downstream statistical analyses like differential abundance testing with tools such as ANCOM-BC or ALDEx2, which rely on accurate count matrices to identify biologically meaningful shifts in community structure.

ASV pipelines incorporate de novo chimera detection as a final quality control step. Chimeras—artifactual sequences formed during PCR from two or more parent templates—are identified by comparing each ASV against more abundant ASVs in the sample. Algorithms flag sequences that can be exactly reconstructed from two parent sequences with a perfect prefix-suffix match. This step removes spurious diversity that would otherwise inflate richness estimates and confound ecological interpretation.

The predecessor to ASVs, OTUs are clusters of sequencing reads grouped by a fixed similarity threshold—typically 97%—as a proxy for species-level distinctions. This approach suffers from several fundamental limitations:

• Arbitrary thresholds collapse true biological variation below the cutoff
• Study-dependent clusters prevent cross-study comparison and meta-analysis
• Loss of single-nucleotide resolution obscures strain-level dynamics ASVs resolve these issues by treating exact sequences as the unit of analysis, enabling reproducible, study-independent definitions that capture the full spectrum of biological variation present in the sample.

Two fundamentally different paradigms for processing amplicon sequencing data:

Denoising (ASV approach)

• Models the error-generating process of sequencing
• Uses quality scores and abundance distributions to distinguish errors from true sequences
• Produces exact, reproducible sequence variants
• Implemented in DADA2, UNOISE3, and Deblur

Clustering (OTU approach)

• Groups reads by pairwise distance into operational units
• Relies on a fixed similarity threshold (e.g., 97%)
• Loses fine-grained biological variation
• Implemented in UPARSE, mothur, and CD-HIT

Denoising has become the preferred method because it captures strain-level variation and produces results that are directly comparable across independent studies.

The most widely used phylogenetic marker for bacterial and archaeal identification in amplicon-based metagenomic studies. This gene contains a mosaic of highly conserved regions interspersed with nine hypervariable regions (V1-V9) that provide taxonomic resolution at different phylogenetic depths. ASV analysis typically targets one or more hypervariable regions—commonly V3-V4 or V4—amplified with universal primers. The conserved flanking regions enable broad-range amplification while the variable regions provide the sequence diversity necessary for resolving taxa to the genus or species level. ASV methods exploit this sequence variation at single-nucleotide precision to distinguish closely related organisms that identical OTU clustering would collapse.

A quantitative measure of the ecological diversity within a single sample, calculated from ASV abundance tables. Key metrics include:

• Observed ASVs: Simple count of unique ASVs (richness)
• Shannon Index: Accounts for both richness and evenness, weighting rare and abundant ASVs
• Chao1 Estimator: Estimates true richness by modeling undetected rare ASVs from singleton and doubleton counts
• Faith's Phylogenetic Diversity: Sums branch lengths of the phylogenetic tree connecting all ASVs in a sample

ASV-level analysis provides more accurate diversity estimates than OTU-based methods because it captures the full spectrum of rare variants that contribute to richness estimates, avoiding the artificial inflation of diversity metrics caused by clustering errors.