This page defines the Cluster Analysis for the PRS. It imports the ClustJS SDK and builds multiple clustering views from window.prsResults.

Clustering views

• PRS clustering — converts flat PRS results into a user × PGS matrix. Rows are users, columns are PGS models, values are PRS scores. Used to cluster users and/or PGS entries.
• Users × one PGS — builds a user × SNP matrix for one selected PGS model. Values can be allele counts or risk × allele. Supports all variants, overlapping variants, and shared variants. Missing variants can be shown as -1 (typically colored black).
• PGS × one user — builds a PGS × SNP matrix for one selected user. Rows are PGS entries, columns are SNPs, values are allele counts. Useful for comparing how multiple PGS models overlap within one person.
• PGS × PGS effect-weight clustering — builds a matrix from PGS scoring-file effect_weight values. Rows are PGS models, columns are SNPs, values are effect weights (often z-scored by row). Compares PGS entries by their SNP effect profiles.
• Genotype-level clustering (E) — builds a users × shared SNPs matrix using only the SNP positions that appear in every loaded user's genome. Values are encoded genotypes. Useful for seeing population-level allele patterns at the positions all users share.
• Genome-wide genotype clustering (F) — clusters users using their raw 23andMe genotype files directly, rather than restricting to SNPs that appear in a PGS model.
  • In PGS-based views the SNP set comes from a selected PGS file (e.g. 77 variants for PGS000001). Section F ignores PGS models entirely and reads across the raw 23andMe files themselves.
  • The matrix is users × SNPs. Values are encoded genotypes — encodeGenotype() maps diploid calls such as AA, AG, or TT to integers 0–9. Missing or invalid calls become -1 (the sentinel value), typically colored black so absent SNPs are immediately visible.
  • Example layout:

              rs123  rs456  rs789  rs999
    user1        0      2     -1      7
    user2        0     -1      4      7
    user3        1      2      4     -1

  • Instead of asking "Are users similar for this one PGS?", Section F asks "Are users similar across many raw 23andMe SNPs?"
  • Useful for clustering by general genotype similarity, platform missingness patterns, or broad ancestry-like structure — not restricted to disease-specific PRS variants.

Session cache

• Stores pivoted PRS matrices, selected PGS/user IDs, allele matrices, PGS-vs-SNP matrices, effect-weight matrices, SNP lists, and raw genotype matrices so they are not recomputed on every re-render.

Helper functions

• Pivot window.prsResults into matrix format.
• Extract unique PGS IDs and user IDs.
• Compute allele-sharing distance between genotypes.
• Encode genotypes (e.g. AA, AG, TT) into numeric values.
• Build raw genotype matrices across users.
• Export matrices as CSV.
• Build matrices for heatmaps and dendrograms.

UI controls

• Row & column clustering toggles.
• Linkage methods: complete, single, average, ward.
• Distance metrics: Euclidean, Manhattan, cosine.
• Dropdowns for selecting PGS models or users.
• Buttons for switching between sections A–F.
• D3 / ClustJS heatmap and dendrogram rendering.

In short: this page turns PRS and genotype-matching results into matrices, then renders interactive clustering views so you can explore patterns across users, PGS models, SNPs, allele counts, and effect weights.