Fig. 3

Sequre’s usability and performance improvements in three biomedical applications. Sequre’s automated code transformation and optimizations reduce A code complexity (number of the lines of code [LOC] in the implementation) and B runtime (execution time in seconds). We show Sequre’s improvements in these metrics in three applications: genome-wide association studies (GWAS), drug-target interaction (DTI) prediction, and metagenomic binning. For metagenomic binning, we consider two recent algorithms, Ganon [22] and Opal [23]. We implemented the analysis pipeline for each application in Sequre and compared with the version without the compiler optimizations as well as implementations in existing frameworks where applicable (C/C++ and PySyft). Note that the C/C++ baselines refer to the recently published, manually optimized MPC implementations of GWAS [9] and DTI prediction [10]. There is no prior MPC implementation for metagenomic binning. Contributions of individual optimization modules in Sequre are shown in different colors within a bar. Sequre generates high-performance MPC programs while allowing them to be easily and compactly written in standard Python language