advanced 8 min read
Life sciences · Topic
Quantitative Trait Loci
probability theory · hypothesis testing · linear algebra · monte carlo methods
Quantitative trait locus (QTL) mapping identifies chromosomal regions that harbour genetic variants affecting a measurable trait. Unlike GWAS which tests individual SNPs, QTL methods leverage linkage information from structured crosses or pedigrees, using interval mapping and LOD score statistics to localise causal regions.

Interval Mapping and the LOD Score

Lander and Botstein (1989) introduced interval mapping, which evaluates QTL presence at each position $\lambda$ along a chromosome using a likelihood ratio:

\[\text{LOD}(\lambda) = \log_{10}\!\left(\frac{L(\text{QTL at }\lambda)}{L(\text{no QTL})}\right)\]

Under the QTL model, phenotype $y_i$ given flanking marker genotypes has a mixture distribution. A LOD $> 3$ (corresponding roughly to $p < 0.001$ genome-wide) is the traditional threshold.

Marker Regression at a Single Marker

The simplest form regresses phenotype on marker genotype class. For an $F_2$ intercross with QTL genotypes $QQ$, $Qq$, $qq$:

Genotype Frequency Mean
$QQ$ $\tfrac{1}{4}$ $\mu + a$
$Qq$ $\tfrac{1}{2}$ $\mu + d$
$qq$ $\tfrac{1}{4}$ $\mu - a$

where $a$ is the additive effect and $d$ the dominance deviation. The $F$-test on the regression recovers the marker–trait association.

Composite Interval Mapping

Composite interval mapping (CIM) conditions on flanking markers to reduce residual variance and control for linked QTL:

\[y_i = \mu + \beta^* x^*_i + \sum_{k \in \mathcal{C}} b_k m_{ik} + \varepsilon_i\]

where $x^*_i$ is the QTL genotype probability at position $\lambda$ (computed from flanking markers via recombination fractions), and the cofactor markers $\mathcal{C}$ absorb variance from elsewhere in the genome.

Permutation Thresholds

Because the LOD profile is correlated across positions, analytical $p$-value thresholds are conservative or anti-conservative. Churchill and Doerge (1994) proposed genome-wide thresholds from permutation:

  1. Permute phenotype–genotype labels (breaks QTL signal, preserves marker correlation).
  2. Record maximum LOD across the genome for each permutation.
  3. Use the 95th percentile of the empirical maximum-LOD distribution as the significance threshold.

Typically $B = 1000$ permutations suffice; this is the gold standard for single-QTL tests.