From Genotype to Phenotype: polygenic prediction of complex human traits (2101.05870v1)

Abstract: Decoding the genome confers the capability to predict characteristics of the organism(phenotype) from DNA (genotype). We describe the present status and future prospects of genomic prediction of complex traits in humans. Some highly heritable complex phenotypes such as height and other quantitative traits can already be predicted with reasonable accuracy from DNA alone. For many diseases, including important common conditions such as coronary artery disease, breast cancer, type I and II diabetes, individuals with outlier polygenic scores (e.g., top few percent) have been shown to have 5 or even 10 times higher risk than average. Several psychiatric conditions such as schizophrenia and autism also fall into this category. We discuss related topics such as the genetic architecture of complex traits, sibling validation of polygenic scores, and applications to adult health, in vitro fertilization (embryo selection), and genetic engineering.

• The paper demonstrates that ML techniques accurately predict traits like height with correlation values of 0.639 using thousands of SNPs.
• The paper highlights the effective use of LASSO regression and sparse learning methods to isolate impactful genetic variants.
• The paper discusses how improved polygenic predictions can enhance personalized medicine while emphasizing the need for diverse training datasets.

Polygenic Prediction of Complex Human Traits: Current State and Future Prospects

The paper “From Genotype to Phenotype: Polygenic Prediction of Complex Human Traits” presents an in-depth examination of the field of genomic prediction, which seeks to map genotypic data to phenotypic expressions such as physical traits and disease risks. The authors, Raben et al., explore the current capabilities and limitations of polygenic prediction models, underscoring the transformative potential of these models in fields like personalized medicine and genetic engineering.

Overview of Findings

The authors begin by detailing the ability of machine learning techniques to predict complex traits such as height and various human health conditions from polygenic risk scores (PRS). With advancements in ML, it is now possible to predict certain quantitative traits with significant accuracy based on DNA alone. For instance, traits such as height and heel bone density are predicted with correlation values of 0.639 and 0.449 respectively, based on thousands of single nucleotide polymorphisms (SNPs).

Significantly, individuals with extreme polygenic scores exhibit much higher risks for certain diseases; for example, individuals in the top percentile of PRS may have up to tenfold increased risks for conditions like coronary artery disease and type 2 diabetes. The paper provides comprehensive statistical data, including Tables and Figures, to demonstrate these prediction capabilities. For example, Figure 1 illustrates how genetic predictors have accurately identified high-risk individuals among tens of thousands of siblings for multiple conditions like coronary artery disease and type 1 diabetes.

Technical Insights

The authors delve into the methodologies used for polygenic prediction, with L1 penalized regression, known as LASSO, being notably effective. With sparse learning methodologies, predictors are formed with an assumption that only a minority of all possible genetic variants significantly affect trait variance. The paper highlights the importance of large training datasets which currently represent a bottleneck in advancing the accuracy of polygenic predictions further. They also discuss the role of lineage-specific genotypic data, pointing out that current models mostly cater to populations of European ancestry, hence the need for broader inclusiveness in future studies.

Implications and Future Directions

The implications of genomic prediction extend beyond predictive healthcare. In vitro fertilization (IVF) and embryo selection stand as prime beneficiaries of polygenic advancements, potentially allowing selection against increased disease risk. The paper also touches on germline editing, although practical applications remain speculative given the current understanding of causal variants versus tags.

As genomic prediction technology continues its trajectory toward capturing more comprehensive heritability estimates of human traits, potential applications in personalized healthcare become more feasible. The vision of a healthcare system offering early interventions and tailored treatments based on individual genetic risk profiles is, according to this research, plausible within the next decade, fundamentally altering the landscape of medical diagnostics and preventive strategies.

Conclusion

This comprehensive review by Raben et al. offers a thorough overview of the current state-of-the-art in genomic prediction and delineates key areas where future research and development must focus. With ethical considerations acknowledged, particularly concerning widespread clinical application and genetic engineering possibilities, the paper provides a clear perspective on the dual-track nature of genomic prediction: it is as much a technological frontier as it is a societal one.