Predicting gene perturbation in a purely data-driven way
One of the key questions that scientists often address is ‘What happens when a system is perturbed in a certain way?’ An international team of researchers, including computer scientist Joris Mooij from the UvA, has now shown that these questions can be answered without having to perform the actual perturbation experiment. The scientific journal PNAS published the findings this week.
As a concrete casus, a team of statisticians from the ETH Zürich and computer scientists from the University of Amsterdam, addressed a biological question: if a certain gene of the yeast organism is knocked out, which other genes most strongly change their expression as a result of that perturbation?
They used a recently developed causal discovery method to infer the strongest gene regulatory relations in a purely data-driven way. They worked with a dataset of large-scale genome-wide gene perturbation experiments that was provided by researchers from the UMC Utrecht. Part of the data was used for making predictions, the remaining part of the data for validating the predictions. The results suggest that prediction and prioritization of future experimental interventions can be improved by using such causal discovery methods.
Applicable in many domains
Assistant professor Joris Mooij of the UvA Informatics Institute: ‘This may be the first time that automated causal discovery has been convincingly demonstrated to work on real-world data. Remarkably, these methods allow one to predict the effects of perturbations of a system without having to perform the experiment. This can be important when the perturbation experiment cannot be performed, for example, because it is too expensive, unethical or technically impossible. As the method itself is purely data-driven and uses no biological knowledge, it can be applied in many other domains as well.’
Nicolai Meinshausen, Alain Hauser, Joris M. Mooij, Jonas Peters, Philip Versteeg, Peter Bühlmann: 'Methods for causal inference from gene perturbation experiments and validation', in Proceedings of the National Academy of Sciences of the United States of America, vol.113, no.27, pp.7361–7368, 2016