Nuisance Variation Research Articles

Insight and inference for DVARS

Estimates of functional connectivity using resting state functional Magnetic Resonance Imaging (rs-fMRI) are acutely sensitive to artifacts and large scale nuisance variation. As a result much effort is dedicated to preprocessing rs-fMRI data and using diagnostic measures to identify bad scans. One such diagnostic measure is DVARS, the spatial root mean square of the data after temporal differencing. A limitation of DVARS however is the lack of concrete interpretation of the absolute values of DVARS, and finding a threshold to distinguish bad scans from good. In this work we describe a sum of squares decomposition of the entire 4D dataset that shows DVARS to be just one of three sources of variation we refer to as D-var (closely linked to DVARS), S-var and E-var. D-var and S-var partition the sum of squares at adjacent time points, while E-var accounts for edge effects; each can be used to make spatial and temporal summary diagnostic measures. Extending the partitioning to global (and non-global) signal leads to a rs-fMRI DSE table, which decomposes the total and global variability into fast (D-var), slow (S-var) and edge (E-var) components. We find expected values for each component under nominal models, showing how D-var (and thus DVARS) scales with overall variability and is diminished by temporal autocorrelation. Finally we propose a null sampling distribution for DVARS-squared and robust methods to estimate this null model, allowing computation of DVARS p-values. We propose that these diagnostic time series, images, p-values and DSE table will provide a succinct summary of the quality of a rs-fMRI dataset that will support comparisons of datasets over preprocessing steps and between subjects.

Semi-Supervised Sparse Representation Based Classification for Face Recognition With Insufficient Labeled Samples.

This paper addresses the problem of face recognition when there is only few, or even only a single, labeled examples of the face that we wish to recognize. Moreover, these examples are typically corrupted by nuisance variables, both linear (i.e., additive nuisance variables, such as bad lighting and wearing of glasses) and non-linear (i.e., non-additive pixel-wise nuisance variables, such as expression changes). The small number of labeled examples means that it is hard to remove these nuisance variables between the training and testing faces to obtain good recognition performance. To address the problem, we propose a method called semi-supervised sparse representation-based classification. This is based on recent work on sparsity, where faces are represented in terms of two dictionaries: a gallery dictionary consisting of one or more examples of each person, and a variation dictionary representing linear nuisance variables (e.g., different lighting conditions and different glasses). The main idea is that: 1) we use the variation dictionary to characterize the linear nuisance variables via the sparsity framework and 2) prototype face images are estimated as a gallery dictionary via a Gaussian mixture model, with mixed labeled and unlabeled samples in a semi-supervised manner, to deal with the non-linear nuisance variations between labeled and unlabeled samples. We have done experiments with insufficient labeled samples, even when there is only a single labeled sample per person. Our results on the AR, Multi-PIE, CAS-PEAL, and LFW databases demonstrate that the proposed method is able to deliver significantly improved performance over existing methods.

Correcting waveform bias using principal component analysis: Applications in multicentre motion analysis studies

Multicentre studies are rare in three dimensional motion analyses due to challenges associated with combining waveform data from different centres. Principal component analysis (PCA) is a statistical technique that can be used to quantify variability in waveform data and identify group differences. A correction technique based on PCA is proposed that can be used in post processing to remove nuisance variation introduced by the differences between centres. Using this technique, the waveform bias that exists between the two datasets is corrected such that the means agree. No information is lost in the individual datasets, but the overall variability in the combined data is reduced. The correction is demonstrated on gait kinematics with synthesized crosstalk and on gait data from knee arthroplasty patients collected in two centres. The induced crosstalk was successfully removed from the knee joint angle data. In the second example, the removal of the nuisance variation due to the multicentre data collection allowed significant differences in implant type to be identified. This PCA-based technique can be used to correct for differences between waveform datasets in post processing and has the potential to enable multicentre motion analysis studies.

MAST: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single-cell RNA sequencing data.

Single-cell transcriptomics reveals gene expression heterogeneity but suffers from stochastic dropout and characteristic bimodal expression distributions in which expression is either strongly non-zero or non-detectable. We propose a two-part, generalized linear model for such bimodal data that parameterizes both of these features. We argue that the cellular detection rate, the fraction of genes expressed in a cell, should be adjusted for as a source of nuisance variation. Our model provides gene set enrichment analysis tailored to single-cell data. It provides insights into how networks of co-expressed genes evolve across an experimental treatment. MAST is available at https://github.com/RGLab/MAST.Electronic supplementary materialThe online version of this article (doi:10.1186/s13059-015-0844-5) contains supplementary material, which is available to authorized users.

Timing of spring surveys for midcontinent sandhill cranes

ABSTRACTThe U.S. Fish and Wildlife Service has used spring aerial surveys to estimate numbers of migrating sandhill cranes (Grus canadensis) staging in the Platte River Valley of Nebraska, USA. Resulting estimates index the abundance of the midcontinent sandhill crane population and inform harvest management decisions. However, annual changes in the index have exceeded biologically plausible changes in population size (>50% of surveys between 1982 and 2013 indicate >±20% change), raising questions about nuisance variation due to factors such as migration chronology. We used locations of cranes marked with very‐high‐frequency transmitters to estimate migration chronology (i.e., proportions of cranes present within the Platte River Valley). We also used roadside surveys to determine the percentage of cranes staging at the Platte River Valley but outside of the survey area when surveys occur. During March 2001–2007, an average of 86% (71–94%; SD = 7%) of marked cranes were present along the Platte River during scheduled survey dates, and 0–11% of cranes that were present along the Platte River were not within the survey boundaries. Timing of the annual survey generally corresponded with presence of the greatest proportion of marked cranes and with least inter‐annual variation; consequently, accuracy of estimates could not have been improved by surveying on different dates. Conducting the survey earlier would miss birds not yet arriving at the staging site; whereas, a later date would occur at a time when a larger portion of birds may have already departed the staging site and when a greater proportion of birds occurred outside of the surveyed area. Index values used to monitor midcontinent sandhill crane abundance vary annually, in part, due to annual variation in migration chronology and to spatial distribution of cranes in the Platte River Valley; therefore, managers should interpret survey results cautiously, with awareness of a continuing need to identify and understand components of variation. Published 2014. This article is a U.S. Government work and is in the public domain in the USA.

Standardizing the intrinsic brain: Towards robust measurement of inter-individual variation in 1000 functional connectomes

As researchers increase their efforts to characterize variations in the functional connectome across studies and individuals, concerns about the many sources of nuisance variation present and their impact on resting state fMRI (R-fMRI) measures continue to grow. Although substantial within-site variation can exist, efforts to aggregate data across multiple sites such as the 1000 Functional Connectomes Project (FCP) and International Neuroimaging Data-sharing Initiative (INDI) datasets amplify these concerns. The present work draws upon standardization approaches commonly used in the microarray gene expression literature, and to a lesser extent recent imaging studies, and compares them with respect to their impact on relationships between common R-fMRI measures and nuisance variables (e.g., imaging site, motion), as well as phenotypic variables of interest (age, sex). Standardization approaches differed with regard to whether they were applied post-hoc vs. during pre-processing, and at the individual vs. group level; additionally they varied in whether they addressed additive effects vs. additive+multiplicative effects, and were parametric vs. non-parametric. While all standardization approaches were effective at reducing undesirable relationships with nuisance variables, post-hoc approaches were generally more effective than global signal regression (GSR). Across approaches, correction for additive effects (global mean) appeared to be more important than for multiplicative effects (global SD) for all R-fMRI measures, with the exception of amplitude of low frequency fluctuations (ALFF). Group-level post-hoc standardizations for mean-centering and variance-standardization were found to be advantageous in their ability to avoid the introduction of artifactual relationships with standardization parameters; though results between individual and group-level post-hoc approaches were highly similar overall. While post-hoc standardization procedures drastically increased test–retest (TRT) reliability for ALFF, modest reductions were observed for other measures after post-hoc standardizations—a phenomena likely attributable to the separation of voxel-wise from global differences among subjects (global mean and SD demonstrated moderate TRT reliability for these measures). Finally, the present work calls into question previous observations of increased anatomical specificity for GSR over mean centering, and draws attention to the near equivalence of global and gray matter signal regression.

The use of a covariate reduces experimental error in nutrient digestion studies in growing pigs1

Covariance analysis limits error, the degree of nuisance variation, and overparameterizing factors to accurately measure treatment effects. Data dealing with growth, carcass composition, and genetics often use covariates in data analysis. In contrast, nutritional studies typically do not. The objectives of this study were to 1) determine the effect of feeding diets containing dehulled, degermed corn, corn-soybean meal, or distillers dried grains with solubles on nutrient digestibility coefficients, 2) evaluate potential interactive effects between initial and final treatment diets on the final treatment diet effects, and 3) determine if initial criterion (digestibility or physiological values) would effectively correct for variation among pigs that could thereby affect final treatment diet digestibility coefficients. Seventy-two crossbred barrows [(Yorkshire × Landrace × Duroc) × Chester White] were randomly assigned to 1 of the 3 diets within initial dietary treatment for Phase-2 (P2; 14 d). Fecal and blood samples were collected after feeding the Phase-1 (P1) diets for 14 d (trial d-14) and on d 28 after feeding the P2 diets for 14 d. Fecal samples were dried and analyzed for C, ether extract, GE, N, NDF, P, and S. Plasma samples were analyzed for plasma urea N and triacylglycerides. Pigs were fed diets that differed widely in CP, NDF, and P, resulting in an overall decrease in C, GE, NDF, N, P, and S digestibility and plasma urea N and triacylglycerides as dietary fiber increased in P1 and P2 (P < 0.10). There were no differences in P2 criteria due to blocking for the P1 diet. There tended (P = 0.10 to 0.20) to be P1 × P2 interactions for NDF and S, indicating that the response of pigs to the P2 diet may depend on the P1 diet. In contrast, when the P1 variable was used as a covariate for P2 data, it was statistically significant for GE, NDF, N, S, and plasma urea N (P < 0.10) whereas C and ether extract showed tendencies but not for P digestibility or plasma triacylglycerides. In conclusion, if initial diets are known, subsequent treatments should be balanced for the initial diet because of potential of initial diet × final diet interactions. If the initial diets are not known, then the initial digestibility coefficient would be effective in reducing the variation associated with subsequently obtained data and should be considered as a covariate in future grower-finisher swine digestibility research.

A Comparative Study of Bottom-Up and Top-Down Approaches to Speaker Diarization

This paper presents a theoretical framework to analyze the relative merits of the two most general, dominant approaches to speaker diarization involving bottom-up and top-down hierarchical clustering. We present an original qualitative comparison which argues how the two approaches are likely to exhibit different behavior in speaker inventory optimization and model training: bottom-up approaches will capture comparatively purer models and will thus be more sensitive to nuisance variation such as that related to the speech content; top-down approaches, in contrast, will produce less discriminative speaker models but, importantly, models which are potentially better normalized against nuisance variation. We report experiments conducted on two standard, single-channel NIST RT evaluation datasets which validate our hypotheses. Results show that competitive performance can be achieved with both bottom-up and top-down approaches (average DERs of 21% and 22%), and that neither approach is superior. Speaker purification, which aims to improve speaker discrimination, gives more consistent improvements with the top-down system than with the bottom-up system (average DERs of 19% and 25%), thereby confirming that the top-down system is less discriminative and that the bottom-up system is less stable. Finally, we report a new combination strategy that exploits the merits of the two approaches. Combination delivers an average DER of 17% and confirms the intrinsic complementary of the two approaches.

Mixed-Effects Statistical Model for Comparative LC−MS Proteomics Studies

Comparing a protein's concentrations across two or more treatments is the focus of many proteomics studies. A frequent source of measurements for these comparisons is a mass spectrometry (MS) analysis of a protein's peptide ions separated by liquid chromatography (LC) following its enzymatic digestion. Alas, LC-MS identification and quantification of equimolar peptides can vary significantly due to their unequal digestion, separation, and ionization. This unequal measurability of peptides, the largest source of LC-MS nuisance variation, stymies confident comparison of a protein's concentration across treatments. Our objective is to introduce a mixed-effects statistical model for comparative LC-MS proteomics studies. We describe LC-MS peptide abundance with a linear model featuring pivotal terms that account for unequal peptide LC-MS measurability. We advance fitting this model to an often incomplete LC-MS data set with REstricted Maximum Likelihood (REML) estimation, producing estimates of model goodness-of-fit, treatment effects, standard errors, confidence intervals, and protein relative concentrations. We illustrate the model with an experiment featuring a known dilution series of a filamentous ascomycete fungus Trichoderma reesei protein mixture. For 781 of the 1546 T. reesei proteins with sufficient data coverage, the fitted mixed-effects models capably described the LC-MS measurements. The LC-MS measurability terms effectively accounted for this major source of uncertainty. Ninety percent of the relative concentration estimates were within 0.5-fold of the true relative concentrations. Akin to the common ratio method, this model also produced biased estimates, albeit less biased. Bias decreased significantly, both absolutely and relative to the ratio method, as the number of observed peptides per protein increased. Mixed-effects statistical modeling offers a flexible, well-established methodology for comparative proteomics studies integrating common experimental designs with LC-MS sample processing plans. It favorably accounts for the unequal LC-MS measurability of peptides and produces informative quantitative comparisons of a protein's concentration across treatments with objective measures of uncertainties.

Median-of-Subsets Normalization of Intensities for cDNA Array Data

cDNA arrays allow quantitative measurement of expression levels for thousands of genes simultaneously. The measurements are affected by many sources of variation, and substantial improvements in the precision of estimated effects accompany adjustments for these effects. Two generic nuisance variations, one associated with the magnitude of expression and the other associated with array location, are common in data from filter arrays. Procedures, like normalization using lowess regression, are effective at reducing variation associated with magnitude, and they have been widely adopted. However, variation associated with location has received less attention. Here, a simple, but effective method based on localized median is expounded for dealing with these nuisance effects, and its properties are discussed. The proposed methodology handles location-dependent variation ("splotches") and magnitude-dependent variation (background and/or saturation) effectively. The procedure is related to lowess when implemented to adjust magnitude-dependent variation, and it performs similarly. The proposed methodology is illustrated with data from the National Center for Toxicological Research (NCTR), where treatment differences in levels of mRNA from rat hepatocytes were assessed using 33P-labeled samples hybridized to cDNA spotted arrays. Normalizing intensities by the median-of-subsets removes systematic variation associated with the location of a gene on the array and/or the level of its expression. This procedure is easy to implement using iteratively reweighted least-squares algorithms. Although less sophisticated than lowess, this procedure works nearly as well for normalizing intensities based upon their magnitude. Unlike lowess, it can adjust for location-dependent effects.

Median-of-Subsets Normalization of Intensities for cDNA Array Data

cDNA arrays allow quantitative measurement of expression levels for thousands of genes simultaneously. The measurements are affected by many sources of variation, and substantial improvements in the precision of estimated effects accompany adjustments for these effects. Two generic nuisance variations, one associated with the magnitude of expression and the other associated with array location, are common in data from filter arrays. Procedures, like normalization using lowess regression, are effective at reducing variation associated with magnitude, and they have been widely adopted. However, variation associated with location has received less attention. Here, a simple, but effective method based on localized median is expounded for dealing with these nuisance effects, and its properties are discussed. The proposed methodology handles location-dependent variation ("splotches") and magnitude-dependent variation (background and/or saturation) effectively. The procedure is related to lowess when implemented to adjust magnitude-dependent variation, and it performs similarly. The proposed methodology is illustrated with data from the National Center for Toxicological Research (NCTR), where treatment differences in levels of mRNA from rat hepatocytes were assessed using 33P-labeled samples hybridized to cDNA spotted arrays. Normalizing intensities by the median-of-subsets removes systematic variation associated with the location of a gene on the array and/or the level of its expression. This procedure is easy to implement using iteratively reweighted least-squares algorithms. Although less sophisticated than lowess, this procedure works nearly as well for normalizing intensities based upon their magnitude. Unlike lowess, it can adjust for location-dependent effects.

Normalizing Counts and Cerebral Blood Flow Intensity in Functional Imaging Studies of the Human Brain

Image intensity normalization is frequently applied to eliminate or adjust for subject or injection global blood flow (gCBF) and other sources of nuisance variation. Normalization has several other positive effects on the analysis of PET images. However, the choice of an intensity normalization technique affects the statistical and psychometric properties of the image data. We compared three normalization procedures, the ratio approach (regional (r)CBF/gCBF), histogram equalization, and ANCOVA, on both PET count and flow data sets. The ratio method presents the proportional increase of regions, the histogram equalization method offers the relative ranking of intensities over the image, and the ANCOVA method provides statistical deviations from an expected linear model of regional values from the subject's gCBF. The original study used 33 normal subjects in a standard subtraction paradigm. The normalization methods were evaluated on their ability to remove extraneous error variation, induce homogeneity of intersubject variation, and remove unwanted dependencies. In general, the normalization modified the subtraction image more than the individual condition images. All three methods worked well at removing the dependency of rCBF on gCBF in count and flow images. For count data, the three methods also reduced the amount of error variation equally well, improving the signal to noise ratio. For flow data, the histogram equalization and ratio methods worked best at reducing statistical error. All three methods dramatically stabilized the variance over the image.