Analyzing large data sets using a computer system

A method for creating a decision model for classifying very large data sets (e.g., greater than about 107 entries) as being of a first state or of a second state different than the first state, using a computer system, comprising:

• obtaining a first set of data images, a plurality of said images each derived from a first state;
• obtaining a second set of data images, a plurality of said images each derived from a second state;
• aligning and normalizing data images in said first set and said second set using an image processing algorithm;
• deriving a differential image mask by comparing images in said first set and said second set thereby determining image pixel locations that are not useful in distinguishing between said first set and said second set;
• applying said differential image mask to generate a filtered first set and a filtered second set of data images;
• wherein a plurality of said images comprise an array of pixel values;
• expressing said filtered first set and second set of data images as a first set and second set of vectors in a basis of most significant principal components of said pixel values' covariance matrix, wherein said matrix expresses covariance between pixels' expression levels across said filtered first set and second set of data images;
• and
• selecting a decision procedure based on the relative distribution and separability of said first set and second set of vectors. and;
• classifying an unknown sample using said model, said classifying comprising:
• obtaining an unclassified data image derived from an unclassified very large sample data set;
• aligning and normalizing said unclassified data image such that said image is comparable with one or more images used to create said model using an image processing algorithm;
• filtering said unclassified data image using said differential image;
• expressing said filtered unclassified data image as a vector in the basis of most significant principal components of said model;
• and
• classifying said vector based on said vector's location (as denoted by its filtered, transformed and abbreviated coordinates) as compared with said first vector set and said second vector set;
• outputting results of said classifying.

A method of analyzing liquid-chromatography-mass-spectroscope (LC-MS) maps using a computer system comprising:

• transforming a plurality of raw LC-MS data sets maps using image processing techniques into a plurality of LC-MS images;
• aligning said LC-MS images in LC retention time (RT) and mass-to-charge ratio (m/z) dimensions;
• normalizing expression intensity levels of said LC-MS images;
• filtering said LC-MS images by constructing and applying a series of binary masks corresponding to differential regions of various statistical significances between target and normal LC-MS images;
• determining locations of target samples using map eigenfeatures;
• differentiating between cancer and normal samples using a decision procedure;
• and
• outputting one or more data signals indicating whether a sample is cancerous or normal.

A bioinformatics pipeline for analyzing proteomics data comprising:

• a quantization logic module performing quantization of detector values into map images;
• an alignment logic module performing alignment of map images;
• a normalization logic module performing normalization of map images;
• a selection logic module for selecting differential signals of map images;
• a statistical data reduction module for simplifying data of map images;
• a discrimination module for discriminating between at least a first state and a second state of map images, and
• a statistical feature detection logic module performing feature detection of map images.