A factorization method for the classification of infrared spectra

Abstract

BackgroundBioinformatics data analysis often deals with additive mixtures of signals for which only class labels are known. Then, the overall goal is to estimate class related signals for data mining purposes. A convenient application is metabolic monitoring of patients using infrared spectroscopy. Within an infrared spectrum each single compound contributes quantitatively to the measurement.ResultsIn this work, we propose a novel factorization technique for additive signal factorization that allows learning from classified samples. We define a composed loss function for this task and analytically derive a closed form equation such that training a model reduces to searching for an optimal threshold vector. Our experiments, carried out on synthetic and clinical data, show a sensitivity of up to 0.958 and specificity of up to 0.841 for a 15-class problem of disease classification. Using class and regression information in parallel, our algorithm outperforms linear SVM for training cases having many classes and few data.ConclusionsThe presented factorization method provides a simple and generative model and, therefore, represents a first step towards predictive factorization methods.

A mixture model with a reference-based automatic selection of components for disease classification from protein and/or gene expression levels

“…As an example, brief description of endocrine signalling system, secreting hormones into a blood stream, is given in [1]. Likewise, reference [21] describes how different organs imprint their substances (metabolites) into a urine sample.…”

“…Bioinformatics data analysis is often based on the use of a linear mixture model (LMM) of a sample [ 1 - 15 ], whereas mixture is composed of components generated by unknown number of interfering sources. As an example, components can be generated during disease progression that causes cancerous cells to produce proteins and/or other molecules that can serve as early indicators (biomarkers) representing disease correlated chemical entities.…”

“…However, an identification of individual components within a sample is complicated by the fact that they can be "buried" within multiple substances. In addition to that, dynamic range of their concentrations can vary even several orders of magnitude [ 16 ], i.e., single components could no longer be recognizable [ 1 ]. Nevertheless, there are the algorithms able to extract either individual components or a group of components with similar concentrations within a sample.…”

“…Thus, it is not clear which of the extracted components are to be retained for further prediction/classification analysis. To this end, several contributions toward solution of this problem have been published in [ 1 - 5 , 8 ]. In [ 1 ], a matrix factorization approach to the decomposition of infrared spectra of a sample is proposed taking into account class labels i.e., the classification phase and the components inference tasks are unified.…”

“…In [ 1 ], a matrix factorization approach to the decomposition of infrared spectra of a sample is proposed taking into account class labels i.e., the classification phase and the components inference tasks are unified. Thus, the concept proposed in [ 1 ] is a classifier specific. It is formulated as the multiclass assignment problem where the number of components equals the number of classes and must be less than the number of samples available.…”