Department of Computational Medicine & Bioinformatics

     I am interested in medical devices and drug discovery. Please see Oskar, and a commentary of the lab research in the inaugural issue of Nature Machine Intelligence, and summary 'The Art of Best-performing Algorithms'.


    Our team currently holds
1. the most accuracy algorithm in detecting Parkinson's Disease (winning algorithm 0.87 in AUC in 2017 DREAM PD Digital Biomarker Challenge vs. second place 0.70 in AUC) by mobile devices (iPhone, iWatch, Android, etc.).
2. the most accurate algorithm in detecting sleep apnea and arousal by ECG, EEG, accelometer, oxygen monitors (0.93 AUC in 2018 PhysioNet Challenge).
3. one of the two most accurate algorithms (tied in performance) in mammography reading (0.86 in AUC, winning algorithm in 2017 DREAM mammography Challenge).
All of our medical device algorithms are by deep learning.


     Our team currently holds
1. the most accurate algorithm in predicting cancer drug synergy, reaching the accuracy level of experimental replicates (0.53 in correlation coefficient compared to experiments vs. 0.53 in correlation coefficient by experimental replicates, winning algorithm in DREAM AstraZeneca Drug Combination challenge).
2. The most accurate algorithm in predicting olfaction response by chemical structure, reaching the accuracy of double tests by individuals.
All of our drug research are by structure predictions.


     Our team has written best-performing algorithms and set the state-of-the-art for many other problems beyond device and drug. This includes problems involving transcription factor prediction (ENCODE DREAM), Biomarker selection, Patient Survival and Outcomes and many more. We have contributed the majority of the best-performing algorithms in DREAM challenges, the largest systems biology benchmark study. I am the sole recipient of the DREAM 'Consistent Best Technical Performer' award, and one of the very few people globally who own multiple gold medals in the annual Data Science Bowl by Kaggle, and the best performer of many other Machine Learning Competitions such as Physionet.

     Beyond deep learning, we also contribute to the development of traditional machine learning, I am the inventor of GuanRank (on survival), adaptive GPR and several other algorithms that are often used as the reference algorithms in benchmark studies/challenges. Relevant algorithms have been published in leading journals such as Science, Nature Methods, Nature Communication, etc.


     My dream is to get my inventions into every home, let us be radiology, medicine, home monitoring devices, or just a drop of perfume. Towards this goal, we are not just pursuing a high accuracy method, but also collaborating with doctors, researchers and industry to use it well.