Genetic perturbation screens combined with data-driven statistics to infer functional genetic interactions are powerful methods for the unbiased, large-scale and quantitative analysis of molecular complexes. These interactions can be inferred from large sets of single gene perturbation screens with a quantitative readout, and from double gene perturbation screens to measure synthetic interactions. However, such approaches have thus far not been able to accurately reveal molecular regulatory circuits involving, for instance, kinase-substrate interactions, because of a poor correlation between their perturbation effects across a large number of measurements. In addition, in human cells, large-scale double gene perturbation screens suffer from poor scalability and are technically challenging. Therefore, most screens have been based on single gene perturbations using, for instance, RNAi, and thus do not reveal functional linkages between cellular processes.
To enable the inference of regulatory interactions, and to enable the inference of interactions from single-gene perturbation screens in human cells, we developed a statistical approach that infers interactions when overall good correlations are found and perturbation effects are strong, but, importantly, also when similarities in subset effects are observed, from which it in addition infers a statistical hierarchy. This method, termed the Hierarchical Interaction Score (HIS) , formalises the following biologically motivated and intuitive principle: gene A, which is a hit in a given set of screens, is inferred to act upstream of gene B if B is a hit with the same sign for an exact subset of those screens, and if there is no gene C with an intermediate subset of hits. This means that genes with broader phenotypes are placed upstream of genes with a subset of those phenotypes. Integrating this across a wide range of thresholds applied to the dataset, we conceived a statistical approach that outperforms correlation-based methods in inferring functional interactions from parallel single-gene perturbation screens in both drosophila and human cells, and performs equally well in inferring functional interactions from synthetic double-gene knockout screens in yeast.
We have created an online resource, www.his2graph.net, that allows calculating HIS scores on various types of datasets and visualising HIS interaction networks, as well as browsing pre-calculated HIS networks on a number of example datasets.
Outline of the HIS algorithm
Current lab members involved:
Some relevant publications from the lab: