PROJECT TITLE :

Ultra High-Dimensional Nonlinear Feature Selection for Big Biological Data - 2018

ABSTRACT:

Machine learning strategies are used to get complex nonlinear relationships in biological and medical data. But, refined learning models are computationally unfeasible for data with many features. Here, we introduce the primary feature choice technique for nonlinear learning problems which will scale up to giant, ultra-high dimensional biological information. A lot of specifically, we scale up the novel Hilbert-Schmidt Independence Criterion Lasso (HSIC Lasso) to handle various features with tens of thousand samples. The proposed technique is guaranteed to seek out an optimal subset of maximally predictive features with minimal redundancy, yielding higher predictive power and improved interpretability. Its effectiveness is demonstrated through applications to classify phenotypes based on module expression in human prostate cancer patients and to detect enzymes among protein structures. We have a tendency to achieve high accuracy with as few as twenty out of 1 million options-a dimensionality reduction of 99.998 percent. Our algorithm will be implemented on commodity cloud computing platforms. The dramatic reduction of options may result in the ever present deployment of subtle prediction models in mobile health care applications.