Navigating the Gene Pool
Understanding how genes interact with each other is of paramount importance in developing better gene therapies. However, current gene screening techniques cannot directly inform us how genes interact with each other. In order to better understand the biological processes of gene interaction, the Laboratory of Intelligent Systems (LIS) at the Ecole Polytechnique Fédérale de Lausanne (EPFL), in collaboration with IBM, MIT and Harvard University, has developed a method to compare predictions and discovered a novel technique to find the best results.
Errors in predicting how the genes interact are frequent and have slowed down progress in the field of gene research. Together with colleagues at the IBM T.J. Watson Research Labs in New York, the LIS organized a competition at the Broad Institute of MIT and Harvard where hundreds of predictions submitted by groups all over the world were compared. The competition allowed the researchers not only to identify the typical types of mistakes made by the scientists, but also to develop a method that gives the best possible prediction of gene interaction by pulling together the "community" of predictions made by different teams. The study shows that these community predictions are accurate, even if many individual predictions are poor.
The result of this work provides a powerful method for obtaining the most accurate prediction of gene interactions and highlights very precise pitfalls of current individual methods that will hopefully result in improved predictions in the future. . An article explaining the findings has been published online March 22, on the website of the Proceedings of the National Academy of Sciences of the United States of America (PNAS).