Predicting phenotypic variation from genotypes, phenotypes and a combination of the two

•Predicting how individuals vary is an important challenge for systems biology.•Making predictions from individual genome sequences is one aspect of this.•However genetic predictions have practical and fundamental limitations.•Intermediate phenotypes can capture both genetic and non-genetic influences.•Clinical prediction models should combine genetics with intermediate phenotype measures.

A central challenge for medicine is to predict disease risk and treatment outcomes for individuals. But what kind of information should be used to make useful predictions in biology? One important cause of phenotypic variation is of course genetics. However genetic predictions have both practical and fundamental limitations: most genetic influences on a trait are usually unknown, and phenotypic variation is not just due to genetics. A pragmatic alternative is to use intermediate phenotypes such as gene expression and other molecular measurements to make predictions about later trait variation such as disease risk. Intermediate phenotypes are useful because they capture both genetic and non-genetic influences on a system, and can reflect both the current state of a system and its history. Here we discuss examples of both genetic and non-genetic approaches to predicting phenotypic variation. Moreover, we argue that it will be by combining these two sources of information — genetics and intermediate molecular phenotypes — that it will be possible to make accurate predictions about variation in many phenotypic traits, even if we will not always mechanistically understand why this is the case. In particular, we encourage the human genetics community to focus more on combining genetics with intermediate phenotypes when attempting to predict clinically relevant traits such as disease risk.

Graphical abstractFigure optionsDownload full-size imageDownload high-quality image (142 K)Download as PowerPoint slide