Thermometers: something for statistical geneticists to think about.

Area of Science:

• Human genetics
• Statistical genetics
• Genomic research

Background:

• Common statistical measures in human genetics include maximized likelihood ratios, LOD scores, and empirical p-values.
• These measures are frequently used to assess statistical evidence for genetic findings, particularly in complex disorders.
• Existing evidence measures possess undesirable properties when applied to complex genetic studies.

Purpose of the Study:

• To critically evaluate the properties of current statistical evidence measures in human genetics.
• To identify deficiencies in existing measures, especially concerning their application to complex disorders.
• To propose a theoretical framework and alternative metric for more robust statistical evidence in gene discovery.

Main Methods:

• Analysis of the behavior of common statistical evidence measures (LOD scores, p-values) with increasing data.
• Theoretical derivation of requirements for ideal statistical evidence measures, using a thermometer analogy.
• Development of a novel evidence metric designed to overcome deficiencies of current methods.

Main Results:

• Demonstrated that current measures like LOD scores can erroneously decrease as more data are collected for a linked locus.
• Identified a violation of fundamental assumptions in standard linkage and association study designs.
• Proposed a set of minimal requirements for effective statistical evidence measures.

Conclusions:

• Current statistical evidence measures in human genetics are inadequate for complex disorders due to counterintuitive behavior with more data.
• A theoretically coherent approach to evidence measurement is needed for reliable gene discovery.
• The proposed alternative metric offers a promising direction for improved identification and characterization of genes contributing to complex diseases.