Catching evolution in the act with the Singleton Density Score

A recent study led by Jonathan K. Pritchard at Stanford University brought a media storm with catchy headlines in both of the flagship scientific outlets Nature and Science News. Aside from highlighting the question of preprints without peer review being covered by popular media, it has also raised attention of the scientific community because of the newly described method for detecting recent selection.
The Singleton Density Score (SDS) is a measure based on the idea that changes in allele frequencies induced by recent selection can be observed in a sample’s genealogy as differences in the branch length distribution.

“The key idea underlying SDS is that recent frequency changes generate differences in the distributions of coalescence times on the two allelic backgrounds.”

The new method uses whole-genome data and looks at the variation around SNPs. Assuming that derived alleles increasing in frequency have shorter branches (and ancestral alleles decreasing in frequency have longer branches), these are expected to have fewer mutations. With SDS, Field et al. look at the distance to the nearest singleton upstream and downstream from each SNP.

“We then use the distributions of distances for each of the three genotypes at the test SNP to compute a maximum likelihood estimate of the log ratio of mean tip-branch lengths for the derived vs. ancestral alleles.”

Field et al. 2016, bioRxiv preprint. DOI:

Field et al. 2016, bioRxiv preprint. DOI:


  • Ignores recombination
  • Bias due to missing singletons
  • Possible effect of variable mutation rate
  • Detects selection only up to 100 generations back
  • Can be influenced by population structure


  • Detects very recent selection
  • Detects polygenic selection
  • The two alleles are controls for each other

“More generally, in SDS the two allelic backgrounds provide a natural control against each other for a variety of possible shortcomings in the data and models. Another strength of SDS compared to earlier methods such as iHS, is that the information comes from a large number of nearly independent tip branches, thus explaining why the statistic is well-behaved under the null.”

Field et al. tested the method on data from the UK10K Project and found signals of selection associated with increased height, blonde hair and blue eyes, milk digestion and immunity genes. Such prime examples are exactly what popular media want, so you can read more on the fancy results part in Scientists track last 2,000 years of British evolution or Humans are still evolving—and we can watch it happen.
Field, Y., et al. (2016) Detection of human adaptation during the past 2,000 years. Preprint at bioRxiv

This entry was posted in methods, mutation, population genetics, selection and tagged , , , . Bookmark the permalink.