Enhancement of Explanation of Stanford WebFEATURE System for Function Prediciton in Protein Structures


Ravnish Narula

Oral Defence Date: 

Friday, May 18, 2012 - 11:00


HH 301


11:00 AM


Professors Dragutin Petkovic, James Wong; and CCLS Research Assoc. Mike Wong


Explanation or transparency of Machine Learning (ML) derived results, e.g. the ability to provide easy to understand and intuitively simple explanations for the predictions made are very important in adoption of such systems in practice, especially in the area of medicine and bioinformatics. Stanford WebFEATURE (http://feature.stanford.edu/webfeature/) is a WWW application to scan protein structures for identifying functional sites. It uses ML to identify these sites and hence its ability to adequately explain its decisions to users in bioinformatics research is very important so they can build confidence in the predictions made. Original WebFEATURE provides an explanation page ("more info" page) which offers very basic explanation of results to the user by simply listing some only very basic statistics. The primary goal of this work, part of joint collaborative effort with Stanford Helix Group, was to design and implement improvements of WebFEATURE explanation page in order to provide richer set of functions for improvement of users. understanding of how the predictions were made. To achieve this, we have worked extensively with Stanford team to derive new functions such as sorting of results and new types of information to be shown, and we created an interactive explanation page with new layout. These changes have been implemented, tested over multiple browsers and approved for immediate integration into new version of WebFEATURE. The secondary goal of this project was to implement general improvements on WebFEATURE infrastructure, and for this a number of infrastructure changes have been implemented, tested and integrated with new version of WebFEATURE. In this project we used modern software engineering methods like agile and scrum project management, and User Centered Design with focus groups at SFSU and Stanford for the development of the user interface of the new explanation page.

Ravnish Narula

e.g. Machine Learning explanation, WebFEATURE