Applications of Machine Learning and High‐Dimensional Visualization in Cancer Detection, Diagnosis, and Management

Abstract

Recent technical advances in combinatorial chemistry, genomics, and proteomics have made available large databases of biological and chemical information that have the potential to dramatically improve our understanding of cancer biology at the molecular level. Such an understanding of cancer biology could have a substantial impact on how we detect, diagnose, and manage cancer cases in the clinical setting. One of the biggest challenges facing clinical oncologists is how to extract clinically useful knowledge from the overwhelming amount of raw molecular data that are currently available. In this paper, we discuss how the exploratory data analysis techniques of machine learning and high-dimensional visualization can be applied to extract clinically useful knowledge from a heterogeneous assortment of molecular data. After an introductory overview of machine learning and visualization techniques, we describe two proprietary algorithms (PURS and RadViz) that we have found to be useful in the exploratory analysis of large biological data sets. We next illustrate, by way of three examples, the applicability of these techniques to cancer detection, diagnosis, and management using three very different types of molecular data. We first discuss the use of our exploratory analysis techniques on proteomic mass spectroscopy data for the detection of ovarian cancer. Next, we discuss the diagnostic use of these techniques on gene expression data to differentiate between squamous and adenocarcinoma of the lung. Finally, we illustrate the use of such techniques in selecting from a database of chemical compounds those most effective in managing patients with melanoma versus leukemia.

Visualization of High-Dimensional Combinatorial Catalysis Data

“…Some features of radial visualization are described in the literature. 1,[14][15][16] Glyph Plot. A glyph plot is a technique to visualize highdimensional multivariate data in 2D or 3D space.…”

Concentric RadViz: Visual Exploration of Multi-task Classification

“…RadViz can map high-dimensional data with thousands of dimensions to the visual space in a very robust manner [2]. Furthermore, RadViz is inherently interactive: the DAs may be moved freely over the circle, thus the mapping can be updated according to user interaction [13].…”

“…In order to solve it, they proposed different methods to measure the dissimilarity between dimensions, then used an ant colony optimization heuristic to solve the resulting TSP. McCarthy et al [2] proposed a t-statistic metric to compute how effectively each dimension discriminates classes, arranging the DAs so as to optimize the metric. Recently, Di Caro et al [15] presented two different formulations to the Dimensional Arrangement Problem which have a higher probability of finding the global optimum than the original formulation.…”

“…In this context, there has been much effort towards joining projection-based methods with machine learning tools [2], mainly for Active Learning (AL) purposes. For instance, Teoh et al [3] developed PaintingClass, a system that projects multidimensional data onto a visual space while providing interactive resources for users to steer the construction of a decision tree, from which one can uncover important information.…”

Distinguishing lung tumours from normal lung based on a small set of genes

“…Because the combinatorial scale of trying all possible gene sets requires significant time and computational power, we used sample gene sets defined by three different gene selection methods, as described before [18]. Gene sets were systematically drawn from the reduced set of 162 most significant differentially expressed gene probes representing 159 unique genes (Supplemental Table I).…”