Simultaneous and exact interval estimates for the contrast of two groups based on an extremely high dimensional variable: application to mass spec data

This paper is about constructing confidence bands around ROC curves. We first introduce to the machine learning community three band-generating methods from the medical field, and evaluate how well they perform. Such confidence bands represent the region where the "true" ROC curve is expected to reside, with the designated confidence level. To assess the containment of the bands we begin with a synthetic world where we know the true ROC curve---specifically, where the class-conditional model scores are normally distributed. The only method that attains reasonable containment out-of-the-box produces non-parametric, "fixed-width" bands (FWBs). Next we move to a context more appropriate for machine learning evaluations: bands that with a certain confidence level will bound the performance of the model on future data. We introduce a correction to account for the larger uncertainty, and the widened FWBs continue to have reasonable containment. Finally, we assess the bands on 10 relatively large benchmark data sets. We conclude by recommending these FWBs, noting that being non-parametric they are especially attractive for machine learning studies, where the score distributions (1) clearly are not normal, and (2) even for the same data set vary substantially from learning method to learning method.

We show that a number of apparently disparate problems, involving distribution approximations in the presence of discontinuities, are actually closely related. One class of such problems involves developing bootstrap approximations to the distribution of a sample mean when the sample includes both ordinal and continuous data. Another class involves smoothing a lattice distribution so as to overcome rounding errors in the normal approximation. A third includes kernel methods for smoothing distribution estimates when constructing confidence bands. Each problem in these classes may be modelled in terms of sampling from a mixture of a continuous and a lattice distribution. We quantify the proportion of the continuous component that is sufficient to smooth away difficulties caused by the lattice part. The proportion is surprisingly small ? it is only a little larger than n_1 logn, where ? denotes sample size. Therefore, very few continuous variables are required in order to render a continuity correction unnecessary. The implications of this result in the problem of sampling both ordinal and continuous data are discussed, and numerical aspects are described through a simulation study. The result is also used to characterise bandwidths that are appropriate for smoothing distribution estimators in the confidence band problem. In this setting an empirical method for bandwidth choice is suggested, and a particularly simple derivation of Edgeworth expansions is given.

Radiation-induced acute intestinal symptoms (RIAISs) are the most frequent complication of radiotherapy that causes great pain and limits the treatment efficacy. The aim of this study was to identify serum biomarkers of RIAISs in cervical cancer patients by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF MS). Serum samples were collected from 66 cervical cancer patients prior to pelvic radiotherapy. In our study, RIAISs occurred in 11 patients. An additional 11 patients without RIAISs were selected as controls, whose age, stage, histological type and treatment methods were matched to RIAISs patients. The 22 sera were subsequently analyzed by SELDI-TOF MS, and the resulting protein profiles were evaluated to identify biomarkers using appropriate bioinformatics tools. Comparing the protein profiles of serum samples from the RIAIS group and the control group, it was found that 22 protein peaks were significantly different (P < 0.05), and six of these peaks with mass-to-charge (m/z) ratios of 7514.9, 4603.94, 6887.41, 2769.21, 3839.72 and 4215.7 were successfully identified. A decision tree model of biomarkers was constructed based on three biomarkers (m/z 1270.88, 1503.23 and 7514.90), which separated RIAIS-affected patients from the control group with an accuracy of 81%. This study suggests that serum proteomic analysis by SELDI-TOF MS can identify cervical cancer patients that are susceptible to RIAISs prior to pelvic radiotherapy.

Seldi-tof MS Profiling of Plasma Proteins in Ovarian Cancer

Distributed feature selection (DFS) strategy for microarray gene expression data to improve the classification performance

Proteomic profiling of ovine serum by SELDI-TOF MS: Optimisation, reproducibility and feasibility of biomarker discovery using routinely collected samples

Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF MS) for determining prognosis in advanced stage hormone relapsing prostate cancer

Post-prostatic massage urine specimens (PMUS) are expected to be rich in proteins originating from the prostatic acini. In this study, we created a PMUS bank consisting of 57 samples obtained from patients with biopsy-proven prostate cancer (PC) and 56 samples from subjects with biopsy-proven benign lesions to analyze protein profiles of PMUS by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF MS). Strong anion-exchange (Q10), weak cation-exchange (CM10) and immobilized metal affinity capture (IMAC30) ProteinChip Arrays were used for protein profiling. In PC samples, single-marker analysis detected 49 mass peaks that were significantly up-regulated and 23 peaks that were significantly down-regulated, compared with peaks obtained from benign lesion samples. To confirm reproducibility we performed additional three rounds of assay using CM10 chip with pH 4.0 binding buffer. Among these significant peaks, a peak of m/z 10788 was significant throughout all 4 rounds of assays. For hierarchical clustering analysis (HCA), we used the 72 peaks which revealed significant differences in single-marker analysis. The heat map discriminated PC from benign lesions with a sensitivity of 91.7% and a specificity of 83.3%. Therefore, SELDI-TOF MS profiling of PMUS can be applied to differentiate patients with PC from cancer-free subjects. However, further investigation is required to verify the usefulness of this method in clinical practice.

Class prediction using “omics” data is playing an increasing role in toxicogenomics, diagnosis/prognosis, and risk assessment. These data are usually noisy and represented by relatively few samples and a very large number of predictor variables (e.g., genes of DNA microarray data or m/z peaks of mass spectrometry data). These characteristics manifest the importance of assessing potential random correlation and overfitting of noise for a classification model based on omics data. We present a novel classification method, decision forest (DF), for class prediction using omics data. DF combines the results of multiple heterogeneous but comparable decision tree (DT) models to produce a consensus prediction. The method is less prone to overfitting of noise and chance correlation. A DF model was developed to predict presence of prostate cancer using a proteomic data set generated from surface-enhanced laser deposition/ionization time-of-flight mass spectrometry (SELDI-TOF MS). The degree of chance correlation and prediction confidence of the model was rigorously assessed by extensive cross-validation and randomization testing. Comparison of model prediction with imposed random correlation demonstrated biologic relevance of the model and the reduction of overfitting in DF. Furthermore, two confidence levels (high and low confidences) were assigned to each prediction, where most misclassifications were associated with the low-confidence region. For the high-confidence prediction, the model achieved 99.2% sensitivity and 98.2% specificity. The model also identified a list of significant peaks that could be useful for biomarker identification. DF should be equally applicable to other omics data such as gene expression data or metabolomic data. The DF algorithm is available upon request.

New global technologies, allowing simultaneous analysis of thousands of genes, proteins, and metabolites (so-called “omics” technologies), are being adopted rapidly by industry, academia, and regulatory agencies. This study evaluated the potential of proteomics in ecotoxicological research (i.e., ecotoxicoproteomics). Filter-feeding mussels (Mytilus edulis) were exposed continuously for 3 wk to oil, or oil spiked with alkylphenols and extra polycyclic aromatic hydrocarbons. The influence of chronic exposure on mussel plasma protein expression was investigated utilizing ProteinChip array technology in combination with surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI TOF MS). Results indicated that exposure to spiked oil had a more significant effect on protein expression in mussels than oil alone. In total, 83 mass peaks (intact or modified proteins/peptides) were significantly altered by spiked oil, while 49 were altered by oil. In exposed organisms, the majority of peaks were upregulated compared to controls (i.e., 69% in oil and 71% in spiked oil). Some peaks (32 in total) were affected by both treatments; however, the degree of response was higher in the spiked oil group for 25 of the 32 commonly affected features. Additionally, certain peaks revealed exposure- or gender-specific responses. Multivariate analysis with regression tree-based methods detected protein patterns associated with exposure that correctly classified masked samples with 90–95% accuracy. Similarly, 92% of females and 85% of males were correctly classified (independent of exposure). Results indicate that proteomics have the potential to make a valuable contribution to environmental monitoring and risk assessment.

BackgroundNetworks are ubiquitous in modern cell biology and physiology. A large literature exists for inferring/proposing biological pathways/networks using statistical or machine learning algorithms. Despite these advances a formal testing procedure for analyzing network-level observations is in need of further development. Comparing the behaviour of a pharmacologically altered pathway to its canonical form is an example of a salient one-sample comparison. Locating which pathways differentiate disease from no-disease phenotype may be recast as a two-sample network inference problem.ResultsWe outline an inferential method for performing one- and two-sample hypothesis tests where the sampling unit is a network and the hypotheses are stated via network model(s). We propose a dissimilarity measure that incorporates nearby neighbour information to contrast one or more networks in a statistical test. We demonstrate and explore the utility of our approach with both simulated and microarray data; random graphs and weighted (partial) correlation networks are used to form network models. Using both a well-known diabetes dataset and an ovarian cancer dataset, the methods outlined here could better elucidate co-regulation changes for one or more pathways between two clinically relevant phenotypes.ConclusionsFormal hypothesis tests for gene- or protein-based networks are a logical progression from existing gene-based and gene-set tests for differential expression. Commensurate with the growing appreciation and development of systems biology, the dissimilarity-based testing methods presented here may allow us to improve our understanding of pathways and other complex regulatory systems. The benefit of our method was illustrated under select scenarios.

Growth curve models and analyses have received considerable attention from a number of authors. Recent papers which reference earlier work are Potthoff and Roy [1964], Elston [1964], Rao [1965], and Krishniaiah [1967]. The main concern of these and earlier papers was to establish a model for growth curves and then to perform tests of hypotheses on the population mean growth curve or to construct confidence intervals or bands for the population mean growth curve. In this paper the model of Rao [1965] is used to construct tolerance bands for a population of individual growth curves. In the model employed each individual in the population has a growth curve whose expectation can be expressed as a polynomial in time. The tolerance bands desired are bands such that a given proportion, 7r, of the individuals in the population have their expected growth curves entirely within the bands. In the case where the parameters of the expected growth curves are unknown, a solution is given which contains a proportion ir of the expected growth curves with confidence level approximately (1 a). These bands would be useful, for example, in determining when to move fish from one hatchery pond to another for larger fish. At any time with any set confidence, limits can be found for the size of 90%, say, of the population. Given a set of tolerance bands based on a large number of fish growth curves, the manager can decide at what time any given run of fish can be expected to have a proportion ir of their numbers in a particular size range. In section 5 a numerical example is presented for a sample of fish with approximate linear growth for the time-span of interest. Consider the model for polynomial growth curves

BackgroundIndividual responses to oxaliplatin (L-OHP)-based chemotherapy remain unpredictable. The objective of our study was to find candidate protein markers for tumor sensitivity to L-OHP from intracellular proteins of human colorectal cancer (CRC) cell lines. We performed expression difference mapping (EDM) analysis of whole cell lysates from 11 human CRC cell lines with different sensitivities to L-OHP by using surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF MS), and identified a candidate protein by liquid chromatography/mass spectrometry ion trap time-of-flight (LCMS-IT-TOF).ResultsOf the qualified mass peaks obtained by EDM analysis, 41 proteins were differentially expressed in 11 human colorectal cancer cell lines. Among these proteins, the peak intensity of 11.1 kDa protein was strongly correlated with the L-OHP sensitivity (50% inhibitory concentrations) (P < 0.001, R2 = 0.80). We identified this protein as Protein S100-A10 (S100A10) by MS/MS ion search using LCMS-IT-TOF. We verified its differential expression and the correlation between S100A10 protein expression levels in drug-untreated CRC cells and their L-OHP sensitivities by Western blot analyses. In addition, S100A10 protein expression levels were not correlated with sensitivity to 5-fluorouracil, suggesting that S100A10 is more specific to L-OHP than to 5-fluorouracil in CRC cells. S100A10 was detected in cell culture supernatant, suggesting secretion out of cells.ConclusionsBy proteomic approaches including SELDI technology, we have demonstrated that intracellular S100A10 protein expression levels in drug-untreated CRC cells differ according to cell lines and are significantly correlated with sensitivity of CRC cells to L-OHP exposure. Our findings provide a new clue to searching predictive markers of the response to L-OHP, suggesting that S100A10 is expected to be one of the candidate protein markers.

This study aimed to diagnose renal allograft dysfunction with specific biomarkers by serum proteomic analysis. Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF MS) and bioinformatics (support vector machine and leave-one cross validation) were used to analyze serum proteome. Enrolled patients included 38 biopsy-proved acute rejection (BPAR), 10 acute tubular necrosis (ATN), 24 subclinical rejection (SCR) and 29 stable control recipients verified by protocol biopsy. A characteristic protein profile can be detected in each renal allograft dysfunction group. BPAR patients were differentiated from stable patients with markers of 9710.1, 4971, 6675.5, 8563.8, 6709.2, 9319 and 4476.7 Da with high sensitivity and specificity. ATN can be clearly distinguished from BPAR and stable control. Subclinical rejection differentiated from stable control with markers of 9193.1, 2759.1, 8464.6 Da. The independent blind test yielded with high specificity and sensitivity for each group. Serum proteome analysis by SELDI-TOF MS combined with bioinformatics in renal allograft dysfunction is valuable and promising. Specific markers were detected in each group. Identification of these proteins may prove useful as diagnostic markers for allograft dysfunction and better to elucidate the mechanism of acute rejection.

This study examined the effect of oral glutamine (Gln) on radiation injury in breast cancer patients undergoing radiation therapy. The radiation injury was evaluated using Radiation Therapy Oncology Group (RTOG) scales. Cosmesis was scored. Blood Gln and glutathione (GSH) levels were determined. Serum protein profiling was determined using surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF MS). Patients receiving Gln scored significantly better in RTOG score than the patients receiving placebo. Cosmetic scores averaged excellent in the Gln group vs fair to good in the placebo group. Blood Gln and GSH levels were significantly higher in the Gln group vs the placebo group. Serum protein profiling with SELDI-TOF MS identified a novel Gln-responsive protein that showed amino acid similarity with myoglobin. These results suggest that Gln is an effective way to reduce radiation morbidity to breast cancer and is associated with the increased expression of a novel serum protein biomarker.