"Machine Learning Approaches to Biological Research: Bioimage Informatics and Beyond"

Robert F. Murphy

External Senior Fellow, FRIAS

Ray and Stephanie Lane Professor of Computational Biology, Carnegie Mellon University

"Machine Learning Approaches to Biological Research: Bioimage Informatics and Beyond"

The practice of biological research has been fundamentally changed by the advent of both genome scale, high throughput technologies and the growing use of computational methods for analysis and modeling of biological data.  Both experimental design and standards for data interpretation increasingly reflect these trends.  While the main focus of both high throughput methods and computational analysis remains on simple methods like microarrays for measuring RNA expression and protein binding, significant work is being done on more complex biological data, such as images.  In addition, the conventional “data analysis” paths which consist of converting large amounts of experimental data into a smaller set of measurements are increasingly being replaced or supplemented by the use of machine learning methods that can automate the process of converting experimental results into testable hypothes or models.  A cutting edge example is the automated creation of predictive models from biological images.  This short course will present recent work from my group and others and include coverage of

• Basic principles and paradigms of supervised and unsupervised machine learning

• Concepts of automated image analysis

• Approaches for creating predictive models from images including

• Models of cytoskeletal dynamics learned from fluorescent speckle microscopy

• Models of subcellular distribution learned from confocal microscope images

• Active learning paradigms for closed loop systems of cycles of experimentation, model refinement and model testing