Predicting Gene-Disease Associations Using Multiple Species Data
Abstract
Correctly predicting genes associated with hereditary diseases is a first step to understanding the molecular mechanisms that lead to these diseases, and in the long run, to developing effective remedies for them. In this work, we combine the power of the functional gene-gene interaction networks with the phenotypic information from multiple species in a walk-based framework, and use a novel machine learning formulation called PU learning to infer the weights for walks; we do so by deriving features from walks in a combined network consisting of all our information sources. We evaluate our methods on a number of diseases downloaded from the Online Mendelian Inheritance in Man (OMIM) project. We demonstrate high recall for known diseases by cross-validation, and show that PU learning based methods using walk-based features outperform a state-of-the-art method that uses a similar walk-based framework.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 20, 2011
- Accession Number
- ADA585568
Entities
People
- Ambuj Tewari
- Edward Marcotte
- Inderjit S. Dhillon
- John O. Woods
- Nagarajan Natarajan
- U. M. Blom
Organizations
- University of Texas at Austin