Skip to content

Glossary

Plain-language definitions of the terms used across eskaks and its output. If a concept in the report or docs is unfamiliar, look here first.

The core idea

  • Codon: a triplet of DNA bases that codes for one amino acid. Genes are read three bases at a time.
  • Synonymous change (S): a DNA change that leaves the amino acid unchanged ("silent"). Selection mostly ignores these, so they accumulate freely.
  • Nonsynonymous change (N): a DNA change that alters the amino acid. Selection acts on these.
  • dN: the rate of nonsynonymous change, per nonsynonymous site.
  • dS: the rate of synonymous change, per synonymous site.
  • dN/dS (ω, Ka/Ks): the ratio of the two, comparing species (fixed differences). < 1 purifying, ≈ 1 neutral, > 1 positive selection.
  • pN / pS / pN/pS: the same idea but from polymorphism within a population (variants still segregating), computed from a VCF instead of an alignment.
  • N sites / S sites: the number of opportunities for nonsynonymous vs synonymous change in a gene, given its codons. Used to normalize the raw counts (so a long gene isn't automatically "more selected" than a short one).
  • πN/πS: pN/pS weighted by each variant's allele frequency (turned on with --af-weighted).

Kinds of selection

  • Purifying (negative) selection: amino-acid changes are harmful and get removed. pN/pS < 1. The normal state of a working gene.
  • Neutral evolution: changes are neither good nor bad. pN/pS ≈ 1.
  • Positive / diversifying selection: amino-acid changes are favoured. pN/pS > 1. Look here for drug targets, antigens, immune-evasion genes.
  • Relaxed constraint: a gene that used to be constrained but no longer is (e.g. a pseudogene).

Inputs

  • FASTA: a text file of sequences. For eskaks fasta it must be codon-aligned: every sequence the same length, in frame, gaps in multiples of 3.
  • VCF: Variant Call Format: the list of positions where your samples differ from the reference (the SNPs).
  • GFF3: an annotation file saying where the genes (CDS features) are on the genome.
  • Reference FASTA: the genome sequence the VCF positions refer to.
  • Contig name: the chromosome/sequence identifier (e.g. chr1, NC_000962.3). It must be identical across the VCF, reference, and GFF3, or eskaks can't line them up.
  • Allele frequency (AF): how common a variant is (0–1). Read from the VCF INFO/AF field.

The statistics

  • Neutrality test: for each gene, a two-sided mid-p binomial test asking "is the nonsynonymous fraction of SNPs different from what neutral evolution predicts (N/(N+S))?" Gives a p-value per gene.
  • mid-p: the correction that makes that discrete test usable. A binomial count is a whole number, so the textbook version (which counts the observed count fully on both sides) is left with far more than the nominal 5% in hand and rarely calls anything on a small gene. Mid-p counts the observed count only half on each side, which centres the p-values on 0.5 under a genuine null instead of ~0.8. The trade: the test is no longer exact, and on very small genes its real error rate can run slightly above the nominal one (see per-gene neutrality test).
  • p-value: the probability of seeing a result this extreme if the gene were neutral. Small = surprising = evidence of selection.
  • Multiple testing: testing thousands of genes produces false positives by chance. Two corrections:
  • FDR / Benjamini-Hochberg (q-value): controls the proportion of false discoveries. More permissive; good for screening. Use this one by default.
  • Bonferroni: controls the chance of any false positive. Stricter; good for a confident shortlist.
  • Genome-wide (pooled) pN/pS: one overall ratio, pooling SNP and site counts across all genes, as a summary of the whole coding genome.
  • Bootstrap 95% CI: a confidence interval on the genome-wide estimate, obtained by resampling genes many times (--bootstrap).
  • Wilson confidence interval: a per-gene confidence interval on pN/pS; if it excludes 1, the gene departs from neutrality.

Advanced terms (in the report / advanced flags)

  • κ (kappa): the transition/transversion rate ratio. Some organisms (like M. tuberculosis) mutate transitions much more often; --kappa corrects the site counts for this so pN/pS isn't biased.
  • McDonald-Kreitman (MK) test: contrasts fixed vs polymorphic changes to detect adaptation; reports the Neutrality Index and α (fraction of adaptive substitutions).
  • Genomic inflation factor (λ): how far the whole set of p-values departs from chance. λ ≫ 1 means many strong signals, real widespread selection, or systematic bias / clonal linkage. λ ≈ 1 is the normal reading, but it is not proof of calibration: the test is discrete, so a genuine null already lands a little below 1 (about 0.90 for genes with a handful of SNPs, about 0.97 once they carry tens). Read λ upwards, as an inflation flag, never downwards as a clean bill of health.
  • Genomic control: an optional correction (--genomic-control) that divides each test statistic by λ, for when the inflation is artefactual.
  • z(N): a standardized measure of how far a gene's observed nonsynonymous count is from expectation; a power-aware effect size.
  • Site frequency spectrum (SFS): the distribution of variants by allele frequency. Purifying selection keeps harmful variants rare, so pN/pS falling as frequency rises is a signature of constraint.
  • PE/PPE, PGRS, IS elements: repetitive, hard-to-map gene families (especially in M. tuberculosis) whose SNP calls are often artefacts. --exclude-repetitive drops them from the pooled estimate and the test.
  • Same-codon SNPs (MNV): two or more SNPs inside one codon, carried by the same genome: one multi-nucleotide change, not two independent substitutions. With per-sample genotypes eskaks enumerates the codons the cohort actually holds (haploid, so two alleles in one sample are one molecule) and scores each against the reference by Nei-Gojobori pathway averaging, so the reported amino-acid change is the one that genome encodes and Nd + Sd still equals the number of nucleotide differences. Without genotypes there is nothing to intersect, so each SNP is classified against the reference codon and the affected codons are only flagged, through the allele-frequency bound. See Same-codon SNPs, and --shared-codons to mark the affected variant rows.
  • Nei-Gojobori pathway averaging: how a codon pair differing at more than one base is split into synonymous and nonsynonymous differences: average over the orders in which the changes could have happened, excluding any order that passes through a stop codon. Used by both eskaks paths, the FASTA dN/dS and the VCF multi-nucleotide changes, so the two cannot disagree.

Special output values

  • NaN: "not a number": the estimate is undefined (e.g. sequences too divergent and the correction saturated, or a gene with no variation). Reported as null in JSON.
  • inf: infinity: e.g. dN > 0 but dS = 0, so the ratio is unbounded. Reported as null in JSON.
  • NA: the value was not computed for this row (e.g. an untested gene's p-value).