I am interested in machine learning, network biology that is applied to clinical outcome prediction. The current machine learning field is dominated by gradient descent-based methods; I am more interested in developing the branch of techniques with matrix manipulation to complement the strength of gradient descent. I have written many top-performing algorithms in the systems biology field, as benchmarked by community-based challenges. Most of them involve a strong component of matrix decomposition, multiplication, or transformation. I consider the gradient descent-derived approach in machine learning today and the matrix-derived approach are the natural descendants of two branches of modern math: calculus and linear algebra, which together will deliver the most powerful and accurate prediction algorithms. I encourage Umich students to take advantage of the courses I am teaching to learn these top-performing algorithms.
I enjoy working with kids and found it a rewarding process to build long-term relationship with a smart student and a sweet family. I have coached high school students who placed first in the Southeast Michigan Science Fair, or entered Intel STS finalist. This is my personal blog.
I wrote the 1st place solutions for the following DREAM challenges: ALL CODE/DESCRIPTION ARE ACCESSIBLE HERE.
|2016||AstraZeneca-Sanger Drug Combination Prediction DREAM Challenge||1a. Predict drug synergy using drug combinational training data, expression, CNV, mutation, drug chemical space||https://www.synapse.org/#!Synapse:syn4231880/wiki/235664|
|2016||AstraZeneca-Sanger Drug Combination Prediction DREAM Challenge||1b. Predict drug synergy using genomic information||https://www.synapse.org/#!Synapse:syn4231880/wiki/235664|
|2016||AstraZeneca-Sanger Drug Combination Prediction DREAM Challenge||2. Predict drug synergy without drug combination training data (novel drug pairs)||https://www.synapse.org/#!Synapse:syn4231880/wiki/235664|
|2016||ICGC-TCGA-DREAM Somatic Mutation Calling Challenge -- Tumor Heterogeneity and Evolution||Inferring tumor heterogeneity and subclone reconstruction||https://www.synapse.org/#!Synapse:syn6087005/wiki/398911|
|2015||ALS Stratification Prize4Life Challenge||1. Predict survival in ALS PROACT data||https://www.synapse.org/#!Synapse:syn2873386/wiki/391432|
|2015||ALS Stratification Prize4Life Challenge||3. Predict survival for national registry data||https://www.synapse.org/#!Synapse:syn2873386/wiki/391432|
|2015||Olfaction Challenge||1. Predict olfaction response for individuals using chemical structure data||https://www.synapse.org/#!Synapse:syn2811262/wiki/78388|
|2015||Prostate Cancer Challenge||2. Predict which patients cannot tolerate chemotherapy||https://www.synapse.org/#!Synapse:syn2813558/wiki/235080|
|2014||Rheumatoid Arthritis Challenge||1. Predict patient response to specific drugs||https://www.synapse.org/#!Synapse:syn1734172/wiki/65262|
|2014||Rheumatoid Arthritis Challenge||2. Predict the patients that do not respond to treatment (non-responders)||https://www.synapse.org/#!Synapse:syn1734172/wiki/65262|
|2014||Alzheimer BigData Challenge #1||1. Predict the prognosis of AD patients.||https://www.synapse.org/#!Synapse:syn2290704/wiki/70719|
|2014||Alzheimer BigData Challenge #1||3. Classify individuals into diagnostic groups using MR imaging.||https://www.synapse.org/#!Synapse:syn2290704/wiki/70719|
|2014||Broad Institute Gene Essentiality Challenge||2. Identify the most predictive biomarkers to predict cancer cell survivability under small molecule perturbations||http://support.sagebase.org/sagebase/topics/broad-dream-gene-essentialit...|
|2013||Breast Cancer Network Inference||2. Drug response of phosphorylation network prediction in cancer cells||https://www.synapse.org/#!Synapse:syn1720047/wiki/60532|