Back

Why linkage disequilibrium measures disagree: Fisher geometry of rare common haplotype structure

Ichikawa, Y.

2026-07-07 genetics
10.64898/2026.07.02.736022 bioRxiv
Show abstract

Conventional LD measures such as r2 perform poorly in the rare common regime, particularly in asymmetric configurations such as nested haplotype structure. Because r2 is symmetric and quadratic, it removes directional structure in two ways: squaring discards the sign, or phase, retained by the signed LD coefficient D, while symmetric normalization hides the asymmetry between the conditional probabilities P(A|B) and P(B|A). Although D recovers the phase, it is locus symmetric and unnormalized; its magnitude is hard to compare across frequency regimes and it does not by itself express which way the asymmetry runs. We therefore analyze the conditional-probability asymmetry {Delta} = P(A|B) - P(B|A), together with r2 and D, as distinct scalar functions on the haplotype simplex under the Fisher information metric. The conditional probabilities P(A|B) and P(B|A) are bounded in [0, 1], directly express carrier-set inclusion, and are more readily visualized than D. Moreover, their difference admits the exact decomposition {Delta} = M + C into a marginal frequency term M and an LD-coupled term C. Prior work has characterized either the mathematical behavior of LD normalizations across allele-frequency space or the Fisher geometry of the haplotype simplex, but not their connection. We bridge this gap by showing that the geometric structure of the simplex explains why LD measures disagree in the rare common regime and why symmetric normalizations such as r2 lose directional information. We show that the fixed-frequency leaf is intrinsically anisotropic, positively curved, and frequency-dependent under the Fisher metric. These geometric predictions are tested empirically , in phased 1000 Genomes data1 and a two locus Wright Fisher model, in a companion paper (Ichikawa, preprint); the present note develops the geometry itself. Keywords: linkage disequilibrium; Fisher information metric; haplotype simplex; rare variant; conditional-probability asymmetry; nested haplotype structure

Matching journals

The top 3 journals account for 50% of the predicted probability mass.

1
GENETICS
483 papers in training set
Top 0.1%
34.2%
2
Theoretical Population Biology
50 papers in training set
Top 0.1%
9.8%
3
Bioinformatics
1204 papers in training set
Top 3%
9.6%
50% of probability mass above
4
Proceedings of the National Academy of Sciences
2444 papers in training set
Top 12%
4.3%
5
Biostatistics
24 papers in training set
Top 0.1%
4.3%
6
Biometrics
23 papers in training set
Top 0.1%
3.2%
7
The American Journal of Human Genetics
234 papers in training set
Top 1%
3.2%
8
PLOS Computational Biology
1863 papers in training set
Top 12%
2.4%
9
PLOS Genetics
862 papers in training set
Top 6%
2.1%
10
Journal of Computational Biology
48 papers in training set
Top 0.7%
1.5%
11
Nature Communications
5641 papers in training set
Top 51%
1.1%
12
Heredity
64 papers in training set
Top 0.8%
1.1%
13
Scientific Reports
3612 papers in training set
Top 65%
1.1%
14
Frontiers in Genetics
230 papers in training set
Top 4%
1.1%
15
G3: Genes, Genomes, Genetics
252 papers in training set
Top 4%
1.0%
16
European Journal of Human Genetics
58 papers in training set
Top 0.9%
1.0%
17
Systematic Biology
144 papers in training set
Top 0.6%
1.0%
18
eLife
5828 papers in training set
Top 65%
0.8%
19
The Annals of Applied Statistics
19 papers in training set
Top 0.2%
0.8%
20
PeerJ
308 papers in training set
Top 11%
0.8%
21
Physical Review Research
49 papers in training set
Top 0.9%
0.8%
22
Genetic Epidemiology
55 papers in training set
Top 0.7%
0.8%
23
Genome Biology
637 papers in training set
Top 9%
0.6%
24
Physical Review Letters
47 papers in training set
Top 0.5%
0.6%