- Optimized the C++ code and running pipeline of the MITE Uncovering SysTem (MUST) to enhance its usability by assuming no prior knowledge of MITEs required from the users - Significantly increased its running speed compared with the first version & the current version also shows significantly increased detection accuracy for recently active MITEs - Prepared a benchmark dataset, the simulated genome with 150MITE copies for interested researchers - Worked as the co-first author and published the paper on the Journal of Integrative Bioinformatics

- Worked as a co-first author in this paper that is under review by Computers in Biology and Medicine - Designed a novel rotation-invariant image feature and Triz demonstrated the effectiveness on both the rotation invariance and the lesion detection of three gastric lesion types - Achieved 87.0% in the four-class classification problems of the three gastric lesion types and the normal controls, averaged over the twenty random runs of 10-fold cross validations

- Worked as a co-first author to completely re-code the previous R program in Python - Optimized the system with parallel computing to improve calculating speed dramatically - Applied statistics in the feature selection to rank the biomedical features with statistical association evaluation algorithm T-test - Compared common machine learning algorithms such as Support Vector Machine (SVM), K Nearest Neighbors (KNN), Decision Tree (DTree) ,Naïve Bayesian classifier (NBayes), Logistic Regression (LR) - Proposed and tested a few novels ideas in the RIFS algorithm which outperforms the existing filter and wrapper feature selection algorithm on both transcriptomic and methylomic datasets - Published the paper on Scientific Reports