Polyhedral Optimizations of RNA-RNA Interaction Predictions-Case Study on Domain-Specific Language

Ribonucleic acid (RNA) is one of the three macromolecules found in mammalian systems. RNA-RNA Interactions (RRI) have been found to aid in Cancer research and treatment. But the problem with the applications that predict these interactions is that they have high computation [θ(N⁶)] and memory costs [θ(N⁴)]. piRNA is one such RRI application. After careful investigation into piRNA, we found that the majority of the computations fit a well-defined mathematical optimization framework called the polyhedral model. But the automatic polyhedral optimization tools and the compilers fail to explore specific optimizations that are known to be within reach of the polyhedral model. This observation facilitated our interest in using a manual optimization tool, AlphaZ. This paper provides the observations made when implementing piRNA application using the domain-specific language called Alpha. This implementation is the first attempt in writing a large real-world application using AlphaZ. For a novice developer, it took about two years to learn the tool, the algorithm and write a functionally correct implementation. We were not successful in applying the optimization techniques to the entire application. But, in our later study, we were able to represent the dominant computation with a surrogate kernel and improve its performance by 100 fold. In this paper, we present the challenges and recommendations for future application implementation using AlphaZ.