Research statement
William Stafford Noble, Assistant Professor
Department of Genome Sciences
Department of Computer Science and Engineering
University of Washington

My research focuses on the development of machine learning techniques for application to problems in molecular biology. I approach these problems using Bayesian techniques such as hidden Markov models, as well as support vector machines and related, non-Bayesian methods. Much of my work addresses two core problems in machine learning: incorporating domain-specific prior knowledge and learning from heterogeneous data. Within computational biology, my research can be divided into six problem domains, as follows:

• Protein homology detection. My work in this field stems from my Ph.D. dissertation, which developed motif-based hidden Markov models (HMMs) of protein families. We have also developed several algorithms that combine results from multiple pairwise sequence comparisons, as well as discriminative classification algorithms based upon the support vector machine (SVM) algorithm.
• Recognition of regulatory modules. Motif-based HMMs can also be used to model collections of transcription factor binding sites in the upstream regions of genes. Identifying co-acting factors and the genes that they operate on is critical to understanding the regulatory machinery of the cell. We have developed two methods for attacking this problem: one that uses the models generatively, and one that uses them in a discriminative framework.
• Microarray expression analysis. During my postdoctoral work with David Haussler, we developed a new method of functionally classifying genes using gene expression data from DNA microarray hybridization experiments. At Columbia, my group developed a collection of complementary microarray analysis techniques. We currently have several ongoing collaborations with microarray researchers to apply and extend our methods.
• Genomic data fusion. Over the next five years, a primary focus of the research in my group will be data fusion. We aim to provide a coherent computational framework for integrating and drawing inferences from a collection of genome-wide measurements. We are focusing on yeast, and we approach this task using two recent trends from the field of machine learning: kernel methods and transduction. This project involves a broad range of collaborators in biology and machine learning.
• Gene finding. In collaboration with David Haussler's group at the University of California, Santa Cruz, we are developing G-Known, a gene-finding system that is designed to be scalable and flexible with respect to the gene features it models, the machine learning algorithms it employs, and the range of experimental data from which it learns.
• Analysis of mass spectrometry data. Recently, in collaboration with Dave Anderson at the University of Oregon and researchers at Rigel, Inc., we have begun working on the problem of peptide identification from protein tandem mass spectra. Our initial work has demonstrated the feasibility of using machine learning techniques in this setting, leading to a significant improvement over previous, non-learning-based methods. We plan to continue this research, developing kernel functions that incorporate prior knowledge about spectra, protein modifications, and an error model.

We are continuing to develop solutions and apply our existing methods in all six of the areas listed above.

Updated March, 2003