Guilt-by-association feature selection: identifying biomarkers from proteomic profiles

DUBStepR is a scalable correlation-based feature selection method for accurately clustering single-cell data

Feature selection (marker gene selection) is widely believed to improve clustering accuracy, and is thus a key component of single cell clustering pipelines. Existing feature selection methods perform inconsistently across datasets, occasionally even resulting in poorer clustering accuracy than without feature selection. Moreover, existing methods ignore information contained in gene-gene correlations. Here, we introduce DUBStepR (Determining the Underlying Basis using Stepwise Regression), a feature selection algorithm that leverages gene-gene correlations with a novel measure of inhomogeneity in feature space, termed the Density Index (DI). Despite selecting a relatively small number of genes, DUBStepR substantially outperformed existing single-cell feature selection methods across diverse clustering benchmarks. Additionally, DUBStepR was the only method to robustly deconvolve T and NK heterogeneity by identifying disease-associated common and rare cell types and subtypes in PBMCs from rheumatoid arthritis patients. DUBStepR is scalable to over a million cells, and can be straightforwardly applied to other data types such as single-cell ATAC-seq. We propose DUBStepR as a general-purpose feature selection solution for accurately clustering single-cell data.

Serial analysis of 38 proteins during the progression of human breast tumor in mice using an antibody colocalization microarray

Proteins in serum or plasma hold great potential for use in disease diagnosis and monitoring. However, the correlation between tumor burden and protein biomarker concentration has not been established. Here, using an antibody colocalization microarray, the protein concentration in serum was measured and compared with the size of mammary xenograft tumors in 11 individual mice from the time of injection; seven blood samples were collected from each tumor-bearing mouse as well as control mice on a weekly basis. The profiles of 38 proteins detected in sera from these animals were analyzed by clustering, and we identified 10 proteins with the greatest relative increase in serum concentration that correlated with growth of the primary mammary tumor. To evaluate the diagnosis of cancer based on these proteins using either an absolute threshold (i.e. a concentration cutoff) or self-referenced differential threshold based on the increase in concentration before cell injection, receiver operating characteristic curves were produced for 10 proteins with increased concentration, and the area under curve was calculated for each time point based on a single protein or on a panel of proteins, in each case showing a rapid increase of the area under curve. Next, the sensitivity and specificity of individual and optimal protein panels were calculated, showing high accuracy as early as week 2. These results provide a foundation for studies of tumor growth through measuring serial changes of protein concentration in animal models.

Protein biomarkers of ovarian cancer: the forest and the trees

The goal of effective population-based screening for ovarian cancer remains elusive despite intense efforts aimed at improving upon biomarker and imaging modalities. While dozens of potential serum biomarkers for ovarian cancer have been identified in recent years, none have yet overcome the limitations that have hindered the clinical use of CA-125. Avenues of opportunity in biomarker development are emerging as investigators are beginning to appreciate the significance of remote, as well as local or regional, sources of biomarkers in the construction of diagnostic panels, as well as the importance of evaluating biomarkers in prediagnostic settings. As the list of candidate biomarkers of ovarian cancer continues to grow, refinements in the methods through which specific proteins are selected for further development as components of diagnostic panels are desperately sought. Such refinements must take into account both the bioinformatic and biological significance of each candidate. Approaches incorporating these considerations may potentially overcome the challenges to early detection posed by the histological heterogeneity of ovarian cancer. Here, we review the recent progress achieved in efforts to develop diagnostic biomarker panels for ovarian cancer and discuss the challenges that remain.

Analysis of biological features associated with meiotic recombination hot and cold spots in Saccharomyces cerevisiae

Meiotic recombination is not distributed uniformly throughout the genome. There are regions of high and low recombination rates called hot and cold spots, respectively. The recombination rate parallels the frequency of DNA double-strand breaks (DSBs) that initiate meiotic recombination. The aim is to identify biological features associated with DSB frequency. We constructed vectors representing various chromatin and sequence-based features for 1179 DSB hot spots and 1028 DSB cold spots. Using a feature selection approach, we have identified five features that distinguish hot from cold spots in Saccharomyces cerevisiae with high accuracy, namely the histone marks H3K4me3, H3K14ac, H3K36me3, and H3K79me3; and GC content. Previous studies have associated H3K4me3, H3K36me3, and GC content with areas of mitotic recombination. H3K14ac and H3K79me3 are novel predictions and thus represent good candidates for further experimental study. We also show nucleosome occupancy maps produced using next generation sequencing exhibit a bias at DSB hot spots and this bias is strong enough to obscure biologically relevant information. A computational approach using feature selection can productively be used to identify promising biological associations. H3K14ac and H3K79me3 are novel predictions of chromatin marks associated with meiotic DSBs. Next generation sequencing can exhibit a bias that is strong enough to lead to incorrect conclusions. Care must be taken when interpreting high throughput sequencing data where systematic biases have been documented.

Peakbin selection in mass spectrometry data using a consensus approach with estimation of distribution algorithms

Progress is continuously being made in the quest for stable biomarkers linked to complex diseases. Mass spectrometers are one of the devices for tackling this problem. The data profiles they produce are noisy and unstable. In these profiles, biomarkers are detected as signal regions (peaks), where control and disease samples behave differently. Mass spectrometry (MS) data generally contain a limited number of samples described by a high number of features. In this work, we present a novel class of evolutionary algorithms, estimation of distribution algorithms (EDA), as an efficient peak selector in this MS domain. There is a trade-of f between the reliability of the detected biomarkers and the low number of samples for analysis. For this reason, we introduce a consensus approach, built upon the classical EDA scheme, that improves stability and robustness of the final set of relevant peaks. An entire data workflow is designed to yield unbiased results. Four publicly available MS data sets (two MALDI-TOF and another two SELDI-TOF) are analyzed. The results are compared to the original works, and a new plot (peak frequential plot) for graphically inspecting the relevant peaks is introduced. A complete online supplementary page, which can be found at http://www.sc.ehu.es/ccwbayes/members/ruben/ms, includes extended info and results, in addition to Matlab scripts and references.

Recent advances in computational analysis of mass spectrometry for proteomic profiling

The proteome, defined as an organism's proteins and their actions, is a highly complex end-effector of molecular and cellular events. Differing amounts of proteins in a sample can be indicators of an individual's health status; thus, it is valuable to identify key proteins that serve as 'biomarkers' for diseases. Since the proteome cannot be simply inferred from the genome due to pre- and posttranslational modifications, a direct approach toward mapping the proteome must be taken. The difficulty in evaluating a large number of individual proteins has been eased with the development of high-throughput methods based on mass spectrometry (MS) of peptide or protein mixtures, bypassing the time-consuming, laborious process of protein purification. However, proteomic profiling by MS requires extensive computational analysis. This article describes key issues and recent advances in computational analysis of mass spectra for biomarker identification.

Bioinformatic-driven search for metabolic biomarkers in disease

The search and validation of novel disease biomarkers requires the complementary power of professional study planning and execution, modern profiling technologies and related bioinformatics tools for data analysis and interpretation. Biomarkers have considerable impact on the care of patients and are urgently needed for advancing diagnostics, prognostics and treatment of disease. This survey article highlights emerging bioinformatics methods for biomarker discovery in clinical metabolomics, focusing on the problem of data preprocessing and consolidation, the data-driven search, verification, prioritization and biological interpretation of putative metabolic candidate biomarkers in disease. In particular, data mining tools suitable for the application to omic data gathered from most frequently-used type of experimental designs, such as case-control or longitudinal biomarker cohort studies, are reviewed and case examples of selected discovery steps are delineated in more detail. This review demonstrates that clinical bioinformatics has evolved into an essential element of biomarker discovery, translating new innovations and successes in profiling technologies and bioinformatics to clinical application.

Demoting redundant features to improve the discriminatory ability in cancer data

The identification of a set of relevant but not redundant features is an important first step in building predictive and diagnostic models from biomedical data sets. Most commonly, individual features are ranked in terms of a quality criterion, out of which the best (first) k features are selected. However, feature ranking methods do not sufficiently account for interactions and correlations between the features. Thus, redundancy is likely to be encountered in the selected features. We present a new algorithm, termed Redundancy Demoting (RD), that takes an arbitrary feature ranking as input, and improves this ranking by identifying redundant features and demoting them to positions in the ranking in which they are not redundant. Redundant features are those that are correlated with other features and not relevant in the sense that they do not improve the discriminatory ability of a set of features. Experiments on two cancer data sets, one melanoma image data set and one lung cancer microarray data set, show that our algorithm greatly improves the feature rankings provided by the methods information gain, ReliefF and Student's t-test in terms of predictive power.

A data-mining approach to biomarker identification from protein profiles using discrete stationary wavelet transform

OBJECTIVE

To develop a new bioinformatic tool based on a data-mining approach for extraction of the most informative proteins that could be used to find the potential biomarkers for the detection of cancer.

METHODS

Two independent datasets from serum samples of 253 ovarian cancer and 167 breast cancer patients were used. The samples were examined by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF MS). The datasets were used to extract the informative proteins using a data-mining method in the discrete stationary wavelet transform domain. As a dimensionality reduction procedure, the hard thresholding method was applied to reduce the number of wavelet coefficients. Also, a distance measure was used to select the most discriminative coefficients. To find the potential biomarkers using the selected wavelet coefficients, we applied the inverse discrete stationary wavelet transform combined with a two-sided t-test.

RESULTS

From the ovarian cancer dataset, a set of five proteins were detected as potential biomarkers that could be used to identify the cancer patients from the healthy cases with accuracy, sensitivity, and specificity of 100%. Also, from the breast cancer dataset, a set of eight proteins were found as the potential biomarkers that could separate the healthy cases from the cancer patients with accuracy of 98.26%, sensitivity of 100%, and specificity of 95.6%.

CONCLUSION

The results have shown that the new bioinformatic tool can be used in combination with the high-throughput proteomic data such as SELDI-TOF MS to find the potential biomarkers with high discriminative power.