Credits

DELPHI is built on datasets and computational tools that are freely available and actively maintained. DELPHI is indebted to these projects, without which the platform would not be possible.

Datasets

  1. The Allen Ancient DNA Resource (AADR) is a comprehensive, curated dataset of ancient genomes with associated metadata, genotyped on a panel of 1.23m variants and updated regularly with newly published genomes. The AADR is the primary dataset behind DELPHI, and is what allows ancient and modern populations to be compared at the same sites.
  2. Poseidon aDNA develops curation standards for ancient DNA and distributes curated genotype and metadata packages. DELPHI reads the AADR through its Poseidon package, which is also the source of the sample metadata shown in the browser.
  3. gnomAD v3.1.2 aggregates and harmonises population-scale genome sequencing data. DELPHI uses its harmonised callset of roughly 66 million sites for the precomputed modern population tracks, which is why modern populations load without any computation at request time.
  4. The Human Genome Diversity Project (HGDP) and the 1000 Genomes Project are the population panels underlying the gnomAD callset. The HGDP sample table is also what lets DELPHI recognise the same individual across gnomAD and the AADR, which name their samples differently, so that a population means the same set of people in both.
  5. GENCODE v19 provides the gene models DELPHI uses for its gene annotation track and for resolving gene names typed into the search box.
  6. CHELSA-TraCE21k provides climate data at 1km and 100-year resolution since the last glacial maximum. DELPHI summarises it into per-sample temperature and precipitation indices, which can be used to sort tracks or to define populations.
  7. The ArchaeoGLOBE project compiles archaeological assessments of regional land use through the Holocene. DELPHI draws its sociocultural variables from it, including the onset of foraging, agriculture, pastoralism and urbanization.
  8. The UK Biobank is the basis of the genetic distance metadata, derived from a principal-component analysis and used to place populations relative to one another.

Computational tools

  1. bed_reader is a Python reader for PLINK filesets with support for cloud-based resources. DELPHI uses it both to pregenerate the modern population tracks and to run a serverless analysis pipeline that reads PLINK files directly from S3, retrieving only the sites a query needs.
  2. NumPy and CuPy carry the numerical work. Population statistics are expressed once as array operations and run unchanged on CPU or GPU, which is what makes it practical to pregenerate whole-genome tracks for every population.
  3. AG Grid renders the sample and population tables. It is what allows the thousands of samples in the browser to be filtered and sorted on any metadata column when assembling a custom population.
Delphi