Book, Thesis, or Chapter Publications

Books

An introductory textbook on learning classifier systems

  • Will N. Browne and Ryan J. Urbanowicz (In Preparation)
  • Springer
    BibTeX
    Not Yet Available

Summary: This books seeks to provide an accessible introduction to learning classifier systems, also known as rule-based machine learning algorithms.  Linked to this book is a basic Python implemenation of a generic Michigan-style supervised learning classifier system which we call the educational LCS (eLCS).  We are shooting to complete the publication of this first LCS textbook, by the end of 2016.

 

Thesis

The detection and characterization of epistasis and heterogeneity: a learning classifier system approach

  • Ryan J. Urbanowicz (2012)
  • PhD Thesis in Genetics at Dartmouth College (Advisor: Jason H. Moore) (Thesis Committee: Mike L. Whitfield, Margaret J. Eppstein, Robert H. Gross, and Tricia A. Thornton-Wells)
    BibTeX
    @book{urbanowicz2012detection,
    title={The detection and characterization of epistasis and heterogeneity: a learning classifier system approach},
    author={Urbanowicz, Ryan John},
    year={2012},
    publisher={Dartmouth College}
    }

Abstract: As the ubiquitous complexity of common disease has become apparent, so has the need for novel tools and strategies that can accommodate complex patterns of association. In particular, the analytic challenges posed by the phenomena known as epistasis and heterogeneity have been largely ignored due to the inherent difficulty of approaching multifactor non-linear relationships. The term epistasis refers to an interaction effect between factors contributing to diesease risk. Heterogeneity refers to the occurrence of similar or identical phenotypes by means of independent contributing factors. Here we focus on the concurrent occurrence of these phenomena as they are likely to appear in studies of common complex disease. In order to address the unique demands of heterogeneity we break from the traditional paradigm of epidemiological modeling wherein the objective is the identification of a single best model describing factors contributing to disease risk. Here we develop, evaluate, and apply a learning classifier system (LCS) algorithm to the identification, modeling, and characterization of susceptibility factors in the concurrent presence of heterogeneity and epistasis. This work includes an examination of existing LCS algorithms, the development of new strategies to address the inherent problem of knowledge discovery in LCSs, the introduction of novel heuristics that improve LCS performance and allow for the explicit characterization of heterogeneity, and an application of these collective advancements to an investigation of bladder cancer susceptibility. Successful analysis of simulated and real-world complex disease associations demonstrates the validity and unique potential of this LCS approach.

Reassessment of a ganglioside-liposome biosensor for the detection of biological toxins

  • Ryan J. Urbanowicz (2005)
  • Masters Dissertation in Biological Engineering at Cornell University (Advisors: Richard A. Durst and Antje J. Baeumner)
    BibTeX
    @phdthesis{urbanowicz2005reassessment,
    title={REASSESSMENT OF A GANGLIOSIDE-LIPOSOME BIOSENSOR FOR THE DETECTION OF BIOLOGICAL TOXINS},
    author={Urbanowicz, Ryan},
    year={2005},
    school={Cornell University}
    }

Abstract: Botulism, the disease caused by the introduction of botulinum toxin (BT) into the body, is a rare but severe ailment that frequently results in death from respiratory failure.  Cholera, the disease caused by the introduction of cholera toxin (CT) into the body, is a much more common problem in third world countries, causing severe diarrhea and dehydration. In this study, the design of previously developed sensitive biosensors for the detection of BT and CT has been scrutinized, with the intention of improving upon the detection limit the BT assay.  This biosensor is a test strip assay which utilizes ganglioside-incorporated liposomes, and toxin antibodies which are immobilized on an analytical zone of a plastic-backed nitrocellulose membrane strip to form a sandwich-type detection mechanism.  The intensity of the band could be visually estimated or measured by densitometry, using computer software.  Previous studies with this design obtained a limit of detection (LoD) of 15 pg/mL and 10 fg/mL in 20 minutes for BT and CT, respectively.  Difficulties in obtaining any concentration gradient in the detection of BT, converted this study into a reassessment of this prior design.  All attempts to obtain a BT biosensor similar to that developed in the previous study failed to obtain a LoD greater than 10 mg/mL, and took at least 35 minutes to complete.  A similar attempt to recreate the CT biosensor design yielded a LoD no greater than 0.64 ng/mL, but took only 15-20 minutes to complete.  The results from this study suggest that the design of the bioassay developed for BT should be reconsidered, and that further studies should be undertaken to consider alternative approaches to this assay.

 

Book Chapters

The rise of genetics-based machine learning in biomedical data mining

  • Ryan J. Urbanowicz (2013)
  • Book Chapter for “Methods in Biomedical Informatics: A Pragmatic Approach” (Ed. Indra Neil Sarkar) Academic Press.
    BibTeX
    @book{sarkar2013methods,
    title={Methods in biomedical informatics: a pragmatic approach},
    author={Sarkar, Indra Neil},
    year={2013},
    publisher={Academic Press}
    }

Abstract: While researchers continue to debate the true nature of common disease etiology, and seek out the underlying biology, the daunting magnitude and complexity of the problem has become clear. As is often said, “the more we learn, the less we know.” This often seems to be true of biomedical studies where technology has paved the way for the age of the “–omes” , and a massive explosion of data collection that, to date, has not proportionally yielded all that much in terms of etiological understanding. Research approaches in the “big data” era present us with an additional set of challenges, including: data management, analysis, and accessibility. Throughout the history of science, having the correct perspective has often proven itself to be the determining factor in our ability to uncover truth. From the shape of the earth, to the cause of infectious disease, having effective tools and an open mind are key to gaining an appropriate perspective.