EQUINE POPULATION GENOMICS / TRAIT CANDIDATES
Speed.
An evolving research panel featuring current candidates. Not all genes in this panel are established to affect speed; genes may be added or removed as evidence changes.
Loading reviewed data…
01 / SNP VARIATION
Variation, gene by gene.
421 samples · 20,115 well-called, variable biallelic SNP positions across the current panel.
Select a SNP for its position and exact counts.
02 / CANDIDATE-PANEL PCA
Across the Speed candidate genes.
421 samples · 7,745 eligible SNPs → 1,002 after LD pruning · same samples and SNPs in both maps.
Gene-balanced comparison
SENSITIVITYSelect a point to inspect its sample record.
Axes stay fixed when filtering. Search highlights samples. Racing Score is an annotation of racing achievement, not a measurement of speed or an association test.
Which genes drive these components?
Contribution is each gene’s share of squared SNP loadings for the selected component. It explains this PCA, not a gene’s effect on speed. Gene balancing gives each gene equal total input variance; it does not give each gene equal influence on every component.
Technical diagnostics
A shared sample set, recorded metadata, and missingness checks support interpretation. Population structure, relatedness and study composition can still influence these descriptive maps.
03 / GENOTYPE HEATMAP
Which samples carry each variant?
421 samples · observed calls only · SNP columns grouped by gene.
Select a cell for the sample, SNP and genotype.
04 / WORKFLOW & DATA
How this panel was made.
Join the reviewed gene matrices by sequencing ID → retain samples meeting QC in every gene → filter common, well-called biallelic SNPs → prune linked markers → fit PCA and a gene-balanced comparison.
Quality rules and cohort accounting
Reuse the existing genotype filters: DP ≥10, GQ ≥20, allele-balance, site-filter and local-context checks. Samples must be the recorded genomic representative and have ≥95% calls across each gene’s frozen SNP panel, with ≥90% confidence-supported gene-body bases already required by the original gene analysis. On this shared cohort, PCA markers require ≥98% calls and minor allele frequency ≥5%. LD pruning removes a later marker when pairwise-complete dosage r² exceeds 0.8 within 1 Mb on the same chromosome.
PCA centers dosage at 2p and scales by √(2p(1−p)); missing calls are replaced by the fitting mean inside PCA only. The comparison scales each pruned gene block to equal total variance. Both models retain up to 10 components. Neither breed nor Racing Score is used to fit the axes.
845 source groups occur in at least one reviewed gene matrix; 421 meet every gene’s PCA criteria and 424 do not. These counts are distinct from the full sequencing census. Raw-only genomes are not included.
| Gene | Local PCA samples | Panel samples | Common SNPs | After LD pruning |
|---|
Candidate evidence and limitations
This panel collects genes selected for ongoing performance research. Evidence ranges from equine trait studies to biological plausibility and preliminary candidate reports. The tag is a research hypothesis, not proof that every gene affects speed. For example, MSTN findings concern race-distance suitability; DMRT3 findings concern gait. Gene pages retain their functional evidence and sources.
Whole gene intervals contain coding and noncoding variants; this is not a coding-only analysis or a complete inventory of all variant types. Multiallelic SNPs, indels and structural changes remain outside this biallelic panel and may be examined in the individual reports. Panel selection, ancestry, relatedness and technical factors can affect clustering. 11 Racing Score matches are descriptive annotations, insufficient to establish a reliable panel–performance relationship here.
PCA and LD-pruning methodology ↗ · DMRT3 gait evidence ↗ · MSTN race-distance evidence ↗
Panel version: 20 October 2026. Candidate membership is versioned; changing a tag requires a new reviewed analysis build.