Modeling the Missing Heritability of Complex Traits

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 2524 papers706 conceptsfrozen 2026-10-04

For decades, a paradox has sat at the heart of human genetics. Twin and family studies consistently show that traits like height, BMI, and psychiatric disorders are highly heritable—often between 50% and 80%. Yet, when researchers scan the entire genome (GWAS) to find the specific variants responsible, the results often explain only a tiny fraction of that variance, sometimes less than 1%.

This "missing heritability" is not just a statistical gap; it represents a fundamental limit in our ability to predict disease and understand biology. Researchers are now moving beyond simple associations between single mutations and broad clinical diagnoses, instead building multi-layered models that integrate intermediate biological signals, environmental context, and refined "endophenotypes."

Refining the Phenotype

One reason genetic signals appear weak is that clinical labels like "Major Depressive Disorder" or "Schizophrenia" are often too broad. They act as umbrella terms for diverse biological realities. By breaking these down into "symptom-level" models, researchers like Jackson G. Thorp have identified novel risk loci that traditional diagnostic categories missed.

A more powerful approach involves "endophenotypes"—measurable biological markers that sit between raw DNA and observable disease. For example, Patrick F. Sullivan and colleagues developed genetic proxies for brain imaging traits, allowing them to estimate brain structure in hundreds of thousands of individuals without an MRI. Similarly, Ziqian Xie used unsupervised deep learning to compress complex brain MRIs into 128-dimensional "latent" traits. These AI-defined phenotypes often have higher heritability than human-defined ones because they capture the specific structural variations that genes actually influence.

The Transcriptomic Bridge

If DNA is the blueprint, RNA is the active construction site. Traditional GWAS looks at the blueprint, but Transcriptome-Wide Association Studies (TWAS) look at the activity. By integrating genetic data with gene expression models, researchers can identify which genes are being "turned up" or "down" in specific tissues.

Bingxin Zhao demonstrated that these transcriptome-based risk scores significantly outperform conventional models in predicting brain phenotypes. This works because a single gene's expression is often influenced by many different genetic variants; aggregating these signals into a single "imputed expression" value reduces noise. Frameworks like SUMMIT further improve this by leveraging massive expression datasets to capture effects in genes with low individual heritability, which are often ignored but can have significant causal impacts on disease.

Context and Interaction

The "missing" heritability also hides in the interaction between nature and nurture. A genetic variant might only increase risk in a specific environment, a phenomenon known as Gene-Environment (GxE) interaction. Felix C. Tropf found that the genetic influence on human fertility is not a constant; it shifts dramatically depending on historical and geographical context.

Similarly, Qiao Fan showed that certain genetic risks for myopia are significantly amplified by higher educational attainment. When models ignore these environmental "triggers," the genetic signal appears diluted. By accounting for education or social deprivation, as explored by Qiang Xu in the context of schizophrenia and depression, researchers can recover some of the variance that traditional models lose.

De-noising Polygenic Risk

Current Polygenic Risk Scores (PRS) are often "blurry" because many disorders share the same genetic roots (pleiotropy). To make these scores clinically useful, researchers are using advanced statistical techniques to "de-noise" the signal. James J. Li used Genomic Structural Equation Modeling to separate the shared genetic risk for ADHD and Autism from the risk unique to each disorder.

Despite these advances, a significant gap remains. Even the most sophisticated models, such as those combining common variants with rare mutations in Tourette Syndrome or Bipolar Disorder, still struggle to reach the predictive power needed for individual-level diagnostics. As Braden Tierney noted, even the "second genome"—the human microbiome—can sometimes outperform the human genome in predicting certain complex diseases. The future of the field lies in integrating these diverse layers—DNA, RNA, environment, and even microbial data—into a unified model of human health.

Papers in this version

Showing 60 of 524 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 432 papers
  2. Version 1 · 2026-10-04 · 524 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_missing_heritability_350277c4,
  title = {missing heritability},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/missing-heritability/rdr_350277c40ecd4abc82a5eaed98703083},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_modeling_the_missing_heritability_of_com_350277c4_v1,
  title = {Modeling the Missing Heritability of Complex Traits},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/modeling-the-missing-heritability-of-complex-traits/rdr_350277c40ecd4abc82a5eaed98703083/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}