Alpha Diversity: Richness and Shannon

Alpha diversity measures how varied the community is inside a single sample, in contrast to beta diversity, which compares communities between samples. For amplicon and metagenomic data it answers two questions at once: how many distinct taxa (features, OTUs, or ASVs) are present, and how evenly the reads are spread across them. Each index below is a different way of collapsing those two ideas into one number per sample.

Richness, Shannon, Simpson

Observed richness is the raw count of features detected in a sample and ignores abundance entirely, so one read and ten thousand reads of a taxon count the same. The Shannon index adds evenness: it is highest when many features occur at similar abundances and drops when a few taxa dominate. The Simpson index reported by these tools is the probability that two reads drawn at random belong to different taxa, so it is driven by the most abundant members and is relatively insensitive to rare ones.

qiime diversity alpha \
  --i-table table.qza \
  --p-metric observed_features \
  --o-alpha-diversity observed_features.qza

qiime diversity alpha \
  --i-table table.qza \
  --p-metric shannon \
  --o-alpha-diversity shannon.qza

qiime diversity alpha \
  --i-table table.qza \
  --p-metric simpson \
  --o-alpha-diversity simpson.qza

Saved SampleData[AlphaDiversity] to: observed_features.qza
Saved SampleData[AlphaDiversity] to: shannon.qza
Saved SampleData[AlphaDiversity] to: simpson.qza

Depth and rarefaction

Richness and Shannon both climb as you sequence deeper, because extra reads keep uncovering rare taxa, so a sample with 50,000 reads looks more diverse than the same community sampled at 5,000. A rarefaction curve plots a diversity metric against subsampled read depth and shows whether it has plateaued, meaning you have sequenced deeply enough to capture most of the richness. To compare samples fairly you either rarefy every sample to a common depth or model depth explicitly; alpha-rarefaction draws the curves and helps you pick that depth.

qiime diversity alpha-rarefaction \
  --i-table table.qza \
  --p-max-depth 20000 \
  --p-metrics observed_features \
  --p-metrics shannon \
  --m-metadata-file metadata.tsv \
  --o-visualization alpha-rarefaction.qzv

library(vegan)

# counts: a samples-by-features matrix of read counts
# meta:   a data frame with a 'group' column, rows aligned to counts

richness <- specnumber(counts)
shannon  <- diversity(counts, index = "shannon")
simpson  <- diversity(counts, index = "simpson")

# rarefy richness to the shallowest sample before comparing
raremax <- min(rowSums(counts))
rarefied_richness <- rarefy(counts, raremax)
rarecurve(counts, step = 500, sample = raremax)

# Shannon is comparatively depth-robust, so we test it across groups;
# rarefy first (as above) if you compare richness instead
kruskal.test(shannon ~ meta$group)

	Kruskal-Wallis rank sum test

data:  shannon by meta$group
Kruskal-Wallis chi-squared = 11.42, df = 2, p-value = 0.003307