Alzheimer's disease classification

As part of my internship at the Agency for Science, Technology and Research (A*STAR), Institute for High Performance Computing (IHPC), I worked under Dr. Wu Yonghui to investigate methods to build classifiers for Alzheimer’s disease based on genetic data.

My internship focused on four different areas:

1. Feature selection
   • Monte Carlo Tree Search (MCTS)
     • MCTS is a best-first tree search algorithm first described by Romaric and Sebag. This can be used to run feature selection, a critical issue in genetic data as the number of features is in the order of millions and there are features with significant correlations.
     • A notebook containing the experiments to run feature selection can be found here.
   • SparSNP
     • a tool used to fit lasso linear models for large SNP datasets, useful also for feature selection
   • Notebook performing experiments with various feature selection methods, including Lasso, MCTS, Variance Inflation Factors (VIFs) and SparSNP can be found here

2. Reinforcement Learning

   a. An attempt was made to use reinforcement learning methods, specifically Deep Q-Networks (DQN) and PPO (proximal policy optimiztaion) to perform classification. The notebook can be found here

   • Can Reinforcement Learning be used in classification.pptx (presentation on using DQN, PPO and MCTS feature selection with comparison to other machine learning techniques)

3. Logic Regression

   • I investigated the use of logic regression, a method for ensembling binary classifiers by Kooperberg and Ruczinski.
   • Monte Carlo Logic Regression was implemented and the notebook can be found here
4. Investigated the ideal ratio of features to samples. Powerpoint presentation can be found here