This repository contains analysis code to accompany the publication 'Harnessing genotype-phenotype nonlinearity to accelerate biological prediction'.
A core focus of genetics is understanding the relationship between genetic variation (genotypes) and biological traits (phenotypes). Many popular models assume that the effects of genotypes on phenotypes are additive and linear. However, non-additive relationships between genes are well known — one gene can influence the effects of another (epistasis), and some genes have multiple phenotypic effects (pleiotropy).
By accounting for such nonlinear interactions between genes and phenotypes, we show that we are able to accurately predict suites of simulated phenotypes. These findings should be of interest to anyone whose work relies on more accurately modeling genotype-phenotype relationships, especially those in the fields of quantitative, population, and human genetics.
Notebooks containing expanded methods and the code + analyses for generating all figures in the publication can be found in 01_code/notebooks/.
01_code/ Contains code for simulating phenotypes, analyzing phenotypes, and predicting phenotypes with an autoencoder.
All data needed to replicate these analyses are available via Zenodo.
Data used to study empirical variation across sets of phenotypes are publicly available and associated with the following publications:
Arabidopsis thaliana: Exposito-Alonso et al. 2019
Saccharomyces cerevisiae: Bloom et al. 2019
Caenorhabditis elegans: Snoek et al. 2019
Mus musculus 1: Gonzales et al. 2018
Mus musculus 2: Bogue et al. 2015
Drosophila melanogaster: Mackay et al. 2012
More details on these datasets are available in the associated pub.
R v4.2.2
ArcadiaColorBrewer v0.0.0.9000
entropy v1.3.1
gplots v3.1.3
inborutils v0.3.0
lmtest v0.9-40
mgcv v1.8-41
missMDA v1.18
RColorBrewer v1.1-3
RNOmni v1.0.1
reticulate v1.27
scales v1.2.1
vioplot v0.4.0
We try to be as open as possible with our work and make all of our code both available and usable. We love receiving feedback at any level, through comments on our pubs or Twitter and issues or pull requests here on GitHub. In turn, we routinely provide public feedback on other people’s work by commenting on preprints, filing issues on repositories when we encounter bugs, and contributing to open-source projects through pull requests and code review.
Anyone is welcome to contribute to our code. When we publish new versions of pubs, we include a link to the "Contributions" page for the relevant GitHub repo in the Acknowledgements/Contributors section. If someone’s contribution has a substantial impact on our scientific direction, the biological result of a project, or the functionality of our code, the pub’s point person may add that person as a formal contributor to the pub with "Critical Feedback" specified as their role.
Our policy is that external contributors cannot be byline-level authors on pubs, simply because we need to ensure that our byline authors are accountable for the quality and integrity of our work, and we must be able to enforce quick turnaround times for internal pub review. We apply this same policy to feedback on the text and other non-code content in pubs.
If you make a substantial contribution, you are welcome to publish it or use it in your own work (in accordance with the license — our pubs are CC BY 4.0 and our code is openly licensed). We encourage anyone to build upon our efforts.