摘要：High throughput biological assays have provided numerous data sources for studying complex interactions between multiple variables in a biological system. Many computational tools for exploring the voluminous biological data are based on pair-wise correlation between variables. Liquid Association (LA) is a novel statistical concept for inferring higher order of association between variables in a system. While LA was originally introduced to study gene-gene interaction involving three genes at a time, it can be applied for correlating biological measurements with clinical variables such as drug sensitivity profiling and patient survival time. It is computationally expensive to compute LA scores for all possible triplets in very large datasets. Here we show how to take advantage of Graphic Processing Units (GPUs) for speeding up the LA computing. Our GPU-accelerated version of LA computation (GALA) achieved nearly 200-fold improvement over the traditional CPU-alone version. A companion package in R was developed for facilitating follow-up analysis and improving user experience. An example on Global Health Observatory data is provided to showcase how LA analysis can be applied in other data intensive fields.
