A recent study describes the development of an atlas of genetic scores that can be used to predict multi-omics traits. The work, published in Nature, highlights the usefulness of using genotype data to predict levels of other biological molecules, combatting the issue of the cost of multi-omics analyses.
Proteomic, transcriptomic and metabolomic data can all be used to tell us one aspect of human health. When used in combination, these approaches can be revolutionary, allowing researchers to elucidate the mechanisms behind complex traits and find potential drug targets. However, “multi-omics” approaches can be costly and time-consuming, meaning this powerful tactic does not always reach its full potential.
The production of most biomolecules – RNA transcripts, proteins and metabolites – is controlled on some level by an individual’s genetic makeup. With this in mind, a team of global researchers developed a machine learning model in the hopes that genotype data could be used to predict other multi-omic traits, for further use in research.
The study utilised data from over 48,000 individuals from the INTERVAL study, for whom both genomic and other multi-omic information was available. The team used this information to train their model to predict the levels of RNA, proteins and metabolites based on relevant genotypes. The model was able to predict levels of over 13,000 RNA transcripts, 2,600 proteins and 800 metabolites using a genetic scoring system. To validate the results, the team used genetic data from the UK Biobank to predict multi-omics traits and any diseases associated with these traits.
The model was tested in seven diverse ancestries. This showed that the use of genetic scoring to predict the presence of other biomolecules is one that can be applied globally. Confident in the robustness of the results, the team created an atlas of genetic scores to predict downstream multi-omic effects, known as OmicsPred.
An open resource
Whilst genomic data is a valuable resource, it is hard to overlook the usefulness of other omics approaches, and the power of these when combined. OmicsPred is the first resource of its kind to allow prediction of these traits on a large scale with only genotypes to hand. The researchers have made the atlas open to all who wish to use it, to allow those who may only have the capacity to perform genetic profiling to use multi-omics data in their research. However, the team stress that the model was trained using data from a very limited background. This could potentially skew the results, but the researchers are keen to train the model using more diverse data.
Overall, the resource should allow for widespread use of multi-omics, hopefully leading to better outcomes in health research. Senior author Michael Inouye stated: “We anticipate the OmicsPred resource will be widely and routinely utilised to investigate multi-omic traits and phenotype associations.”