Cancer drug activity detector* To improve the cost-efficiency and accelerate drug development process, built a Drug Activity Detector to predict drugs' anti-cancer activity based on their structural properties.* Applied Butina clustering algorithm to select 60,000 chemicals from a pool of 300,000 for drug activity test. Reduced the number of labor hours required for drug testing by 80% and saved $100,000.* Built a LightGBM classifier with drug activity test data. Performed hyperparameter tuning to achieve aprecision score of 0.88. Optimized drugs based on insights drawn from analyzing model's feature importance.* Obtained 3 new drugs in 6 months and improved drug discovery rate by 6-fold. The novel model greatlyimproved institutional cost-efficiency on drug development by reducing the project management life cycle from 6-7 years to 3-4 years and reducing the cost from 1 million to half a million.

Lung cancer drug nominator* To facilitate lung cancer diagnosis and treatment, build a Nominator to recommend drugs to patients based on the results from their diagnostic markers testing.* Applied Affinity Propagation Clustering (APC) to select 12 representative cancer subtypes from a panel of 100 lung cancer variants. Identified 220 drugs selectively killing these 12 lung cancer variants from a collection of 300,000 chemicals.* Combined Elastic Net regression and bootstrap resampling to select the most important diagnostic markers that predict a cancer variant's drug sensitivity. Built a Random Forest classifier with selected markers. This lung cancer drug Nominator predicts a cancer patient's drug effectiveness with a precision score of 0.82.* Our pipeline provides a wealth of disease insights and actionable solutions to personalized cancer treatment, and we have published result in the topmost prestigious journal Cell.

*A preclinical anti-cancer trial using the FDA approved drug ABT-199 (venetoclax) in a mouse model of Hedgehog pathway-driven rhabdomyosarcoma.*Optimization of CRISPR-Cas9 technique against undesirable outcomes associated with sgRNA design.*Construction of a new genome-scale lentiviral expression cDNA library suitable for high throughput gain-of-function screens.*Discovery of novel cross-talks between two therapeutically important cellular processes, cell death and Hedgehog signaling using genome scale high throughput cDNA screens.*Identification and characterization of a new class of tankyrase inhibitors that could be used to target Wnt signaling related cancers.*Discovery of new associations between disease related genes and two cancer driven signaling pathways, Hedgehog and Wnt using high throughput siRNA screens. *Establishment of a multiplex luciferase screening platform to explore therapeutic vulnerabilities using small molecule libraries.