Scientific software is widely used in science and engineering. In addition, results obtained from scientific software are used as evidence in research publications. Despite the critical usage of such software, many studies have pointed out a lack of systematic testing of scientific software. As a result, subtle program errors can remain undetected. There are numerous reports of subtle faults in scientific software causing losses of billions of dollars and the withdrawal of scientific publications. This research aims to develop automated techniques for test oracle creation, test case selection, and develop methods for test oracle prioritization targeting scientific software. The intellectual merits of this research are the following: (1) It advances the understanding of the scientific software development process and investigates methods to incorporate systematic software testing activities into scientific software development without interfering with the scientific inquiry, (2) It forges new approaches to develop automated test oracles for programs that produce complex outputs and for programs that produce outputs that are previously unknown, (3) It develops new metrics to measure the effectiveness of partial test oracles and uses them for test oracle prioritization, (4) It extends the boundaries of current test case selection to effectively work with partial or approximate test oracles.
Recent advances in machine learning (ML) has led to its use in safety-critical applications in healthcare. For example, ML is used for predicting drug responses in cancer treatment, driving precision medicine-based treatments for cancer patients, and is used for biological threat detection using large data sets on the order of millions of data points. With the use of ML in such safety-critical systems, assuring their quality becomes extremely important. To this end, we propose to develop resources for the effective utilization of metamorphic testing for the systematic testing of ML applications.
Most of the current research in Bioinformatics is directed toward developing computational models, but relatively little time and effort are spent on developing the corresponding software and on assuring their quality. Therefore, a group of researchers from MSU, and I seek to develop novel tools to conduct systematic testing on protein function annotation (PFA) software.