Computational biology of blood disorders: Patterns and algorithms

The morphological and molecular characterization and of malignant blood disorders is pivotal to ensure a correct diagnosis, guide therapy, and, in the longer term, to identify their molecular causes. Over the past several years, new technology has increased our abilities to describe, typecast, and understand hematologic malignancies beyond what was previously possible. This thesis concerns two such technologies -- microarray-based gene expression profiling and image analysis-based cytomorphology. The overall aim of the thesis is to identify and alleviate specific algorithmic bottlenecks in order to extend the applicability of these two techniques across a range of clinical and experimental settings. Article I concerns image analysis-based hematologic cytomorphology. A new method for locating white blood cells in microscopic images of blood and bone marrow smears is presented. Unlike previous methods, the proposed method automatically separates complex clusters of white blood cells. Article II to IV concern microarray-based gene expression profiling. Article II presents a new strategy for implementing microarray-based subtyping of hematologic disorders in clinical routine. This strategy (cross-platform classification) removes the need for resource-consuming prior acquisition of training data, and hence facilitates the use of gene expression profiling as a diagnostic tool. Proof-of-concept for the feasibility of cross-platform subtyping of acute leukemia is provided. Article III represents a detailed study of ?ontological analysis?, a widely used computational methodology for interpreting results from genome-wide gene expression studies. A set of enhanced computational ontological analysis methods are developed and systematically evaluated. The proposed methods are shown to constitute attractive alternatives to existing ontological analysis methods approaches in that they are threshold-free and have higher statistical powers. The methods are also used for functional profiling of the imatinib-induced transcriptional response in chronic myeloid leukemia cell lines. Finally, Article IV describes improved methods for detecting and delineating genomic regions with aberrant gene expression in cancer gene expression maps. The developed methods are systematically evaluated, and shown to produce comparatively clean segmentation results. The methods are also applied to gene expression profiles from acute leukemia, and are shown to be able to identify genomic regions corresponding to known imbalanced chromosomal aberrations.