Non-linear shrinking of linear model errors.

BACKGROUND: Artificial neural networks (ANNs) can be a powerful tool for spectroscopic data analysis. Their ability to detect and model complex relations in the data may lead to outstanding predictive capabilities, but the predictions themselves are difficult to interpret due to the lack of understanding of the black box ANN models. ANNs and linear methods can be combined by first fitting a linear model to the data followed by a non-linear fitting of the linear model residuals using an ANN. This paper explores the use of residual modelling in high-dimensional data using modern neural network architectures. RESULTS: By combining linear- and ANN modelling, we demonstrate that it is possible to achieve both good model performance while retaining interpretations from the linear part of the model. The proposed residual modelling approach is evaluated on four high-dimensional datasets, representing two regression and two classification problems. Additionally, a demonstration of possible interpretation techniques are included for all datasets. The study concludes that if the modelling problem contains sufficiently complex data (i.e., non-linearities), the residual modelling can in fact improve the performance of a linear model and achieve similar performance as pure ANN models while retaining valuable interpretations for a large proportion of the variance accounted for. SIGNIFICANCE AND NOVELTY: The paper presents a residual modelling scheme using modern neural network architectures. Furthermore, two novel extensions of residual modelling for classification tasks are proposed. The study is seen as a step towards explainable AI, with the aim of making data modelling using artificial neural networks more transparent.

A comparison of methods for fully automatic segmentation of tumors and involved nodes in PET/CT of head and neck cancers.

Target volume delineation is a vital but time-consuming and challenging part of radiotherapy, where the goal is to deliver sufficient dose to the target while reducing risks of side effects. For head and neck cancer (HNC) this is complicated by the complex anatomy of the head and neck region and the proximity of target volumes to organs at risk. The purpose of this study was to compare and evaluate conventional PET thresholding methods, six classical machine learning algorithms and a 2D U-Net convolutional neural network (CNN) for automatic gross tumor volume (GTV) segmentation of HNC in PET/CT images. For the latter two approaches the impact of single versus multimodality input on segmentation quality was also assessed. 197 patients were included in the study. The cohort was split into training and test sets (157 and 40 patients, respectively). Five-fold cross-validation was used on the training set for model comparison and selection. Manual GTV delineations represented the ground truth. Tresholding, classical machine learning and CNN segmentation models were ranked separately according to the cross-validation Sørensen-Dice similarity coefficient (Dice). PET thresholding gave a maximum mean Dice of 0.62, whereas classical machine learning resulted in maximum mean Dice scores of 0.24 (CT) and 0.66 (PET; PET/CT). CNN models obtained maximum mean Dice scores of 0.66 (CT), 0.68 (PET) and 0.74 (PET/CT). The difference in cross-validation Dice between multimodality PET/CT and single modality CNN models was significant (p ≤ 0.0001). The top-ranked PET/CT-based CNN model outperformed the best-performing thresholding and classical machine learning models, giving significantly better segmentations in terms of cross-validation and test set Dice, true positive rate, positive predictive value and surface distance-based metrics (p ≤ 0.0001). Thus, deep learning based on multimodality PET/CT input resulted in superior target coverage and less inclusion of surrounding normal tissue.

Large-scale genomic prediction using singular value decomposition of the genotype matrix.

BACKGROUND: For marker effect models and genomic animal models, computational requirements increase with the number of loci and the number of genotyped individuals, respectively. In the latter case, the inverse genomic relationship matrix (GRM) is typically needed, which is computationally demanding to compute for large datasets. Thus, there is a great need for dimensionality-reduction methods that can analyze massive genomic data. For this purpose, we developed reduced-dimension singular value decomposition (SVD) based models for genomic prediction. METHODS: Fast SVD is performed by analyzing different chromosomes/genome segments in parallel and/or by restricting SVD to a limited core of genotyped individuals, producing chromosome- or segment-specific principal components (PC). Given a limited effective population size, nearly all the genetic variation can be effectively captured by a limited number of PC. Genomic prediction can then be performed either by PC ridge regression (PCRR) or by genomic animal models using an inverse GRM computed from the chosen PC (PCIG). In the latter case, computation of the inverse GRM will be feasible for any number of genotyped individuals and can be readily produced row- or element-wise. RESULTS: Using simulated data, we show that PCRR and PCIG models, using chromosome-wise SVD of a core sample of individuals, are appropriate for genomic prediction in a larger population, and results in virtually identical predicted breeding values as the original full-dimension genomic model (r = 1.000). Compared with other algorithms (e.g. algorithm for proven and young animals, APY), the (chromosome-wise SVD-based) PCRR and PCIG models were more robust to size of the core sample, giving nearly identical results even down to 500 core individuals. The method was also successfully tested on a large multi-breed dataset. CONCLUSIONS: SVD can be used for dimensionality reduction of large genomic datasets. After SVD, genomic prediction using dense genomic data and many genotyped individuals can be done in a computationally efficient manner. Using this method, the resulting genomic estimated breeding values were virtually identical to those computed from a full-dimension genomic model.

Variable selection models for genomic selection using whole-genome sequence data and singular value decomposition.

BACKGROUND: Non-linear Bayesian genomic prediction models such as BayesA/B/C/R involve iteration and mostly Markov chain Monte Carlo (MCMC) algorithms, which are computationally expensive, especially when whole-genome sequence (WGS) data are analyzed. Singular value decomposition (SVD) of the genotype matrix can facilitate genomic prediction in large datasets, and can be used to estimate marker effects and their prediction error variances (PEV) in a computationally efficient manner. Here, we developed, implemented, and evaluated a direct, non-iterative method for the estimation of marker effects for the BayesC genomic prediction model. METHODS: The BayesC model assumes a priori that markers have normally distributed effects with probability [Formula: see text] and no effect with probability (1 - [Formula: see text]). Marker effects and their PEV are estimated by using SVD and the posterior probability of the marker having a non-zero effect is calculated. These posterior probabilities are used to obtain marker-specific effect variances, which are subsequently used to approximate BayesC estimates of marker effects in a linear model. A computer simulation study was conducted to compare alternative genomic prediction methods, where a single reference generation was used to estimate marker effects, which were subsequently used for 10 generations of forward prediction, for which accuracies were evaluated. RESULTS: SVD-based posterior probabilities of markers having non-zero effects were generally lower than MCMC-based posterior probabilities, but for some regions the opposite occurred, resulting in clear signals for QTL-rich regions. The accuracies of breeding values estimated using SVD- and MCMC-based BayesC analyses were similar across the 10 generations of forward prediction. For an intermediate number of generations (2 to 5) of forward prediction, accuracies obtained with the BayesC model tended to be slightly higher than accuracies obtained using the best linear unbiased prediction of SNP effects (SNP-BLUP model). When reducing marker density from WGS data to 30 K, SNP-BLUP tended to yield the highest accuracies, at least in the short term. CONCLUSIONS: Based on SVD of the genotype matrix, we developed a direct method for the calculation of BayesC estimates of marker effects. Although SVD- and MCMC-based marker effects differed slightly, their prediction accuracies were similar. Assuming that the SVD of the marker genotype matrix is already performed for other reasons (e.g. for SNP-BLUP), computation times for the BayesC predictions were comparable to those of SNP-BLUP.

Autodelineation of cervical cancers using multiparametric magnetic resonance imaging and machine learning.

BACKGROUND: Tumour delineation is a challenging, time-consuming and complex part of radiotherapy planning. In this study, an automatic method for delineating locally advanced cervical cancers was developed using a machine learning approach. MATERIALS AND METHODS: A method for tumour segmentation based on image voxel classification using Fisher?s Linear Discriminant Analysis (LDA) was developed. This was applied to magnetic resonance (MR) images of 78 patients with locally advanced cervical cancer. The segmentation was based on multiparametric MRI consisting of T2- weighted (T2w), T1-weighted (T1w) and dynamic contrast-enhanced (DCE) sequences, and included intensity and spatial information from the images. The model was trained and assessed using delineations made by two radiologists. RESULTS: Segmentation based on T2w or T1w images resulted in mean sensitivity and specificity of 94% and 52%, respectively. Including DCE-MR images improved the segmentation model?s performance significantly, giving mean sensitivity and specificity of 85?93%. Comparisons with radiologists? tumour delineations gave Dice similarity coefficients of up to 0.44. CONCLUSION: Voxel classification using a machine learning approach is a flexible and fully automatic method for tumour delineation. Combining all relevant MR image series resulted in high sensitivity and specificity. Moreover, the presented method can be extended to include additional imaging modalities.

A model to secure a stable iodine concentration in milk.

BACKGROUND: Dairy products account for approximately 60% of the iodine intake in the Norwegian population. The iodine concentration in cow's milk varies considerably, depending on feeding practices, season, and amount of iodine and rapeseed products in cow fodder. The variation in iodine in milk affects the risk of iodine deficiency or excess in the population. OBJECTIVE: The first goal of this study was to develop a model to predict the iodine concentration in milk based on the concentration of iodine and rapeseed or glucosinolate in feed, as a tool to securing stable iodine concentration in milk. A second aim was to estimate the impact of different iodine levels in milk on iodine nutrition in the Norwegian population. DESIGN: Two models were developed on the basis of results from eight published and two unpublished studies from the past 20 years. The models were based on different iodine concentrations in the fodder combined with either glucosinolate (Model 1) or rapeseed cake/meal (Model 2). To illustrate the impact of different iodine concentrations in milk on iodine intake, we simulated the iodine contribution from dairy products in different population groups based on food intake data in the most recent dietary surveys in Norway. RESULTS: The models developed could predict iodine concentration in milk. Cross-validation showed good fit and confirmed the explanatory power of the models. Our calculations showed that dairy products with current iodine level in milk (200 µg/kg) cover 68, 49, 108 and 56% of the daily iodine requirements for men, women, 2-year-old children, and pregnant women, respectively. CONCLUSIONS: Securing a stable level of iodine in milk by adjusting iodine concentration in different cow feeds is thus important for preventing excess intake in small children and iodine deficiency in pregnant and non-pregnant women.

Classification of dynamic contrast enhanced MR images of cervical cancers using texture analysis and support vector machines.

Dynamic contrast enhanced MRI (DCE-MRI) provides insight into the vascular properties of tissue. Pharmacokinetic models may be fitted to DCE-MRI uptake patterns, enabling biologically relevant interpretations. The aim of our study was to determine whether treatment outcome for 81 patients with locally advanced cervical cancer could be predicted from parameters of the Brix pharmacokinetic model derived from pre-chemoradiotherapy DCE-MRI. First-order statistical features of the Brix parameters were used. In addition, texture analysis of Brix parameter maps was done by constructing gray level co-occurrence matrices (GLCM) from the maps. Clinical factors and first- and second-order features were used as explanatory variables for support vector machine (SVM) classification, with treatment outcome as response. Classification models were validated using leave-one-out cross-model validation. A random value permutation test was used to evaluate model significance. Features derived from first-order statistics could not discriminate between cured and relapsed patients (specificity 0%-20%, p-values close to unity). However, second-order GLCM features could significantly predict treatment outcome with accuracies (~70%) similar to the clinical factors tumor volume and stage (69%). The results indicate that the spatial relations within the tumor, quantified by texture features, were more suitable for outcome prediction than first-order features.

Is the brain water channel aquaporin-4 a pathogenetic factor in idiopathic intracranial hypertension? Results from a combined clinical and genetic study in a Norwegian cohort.

PURPOSE: Idiopathic intracranial hypertension (IIH) is a condition of increased intracranial pressure of unknown aetiology. Patients with IIH usually suffer from headache and visual disturbances. High intracranial pressure despite normal ventricle size and negative MRI indicate perturbed water flux across cellular membranes, which is provided by the brain water channel aquaporin-4 (AQP4). IIH could be associated with malfunctioning intracerebral water homeostasis and cerebrospinal fluid (CSF) reabsorption based on functional or regulatory alterations of AQP4. METHODS: Clinical data, blood and CSF samples were collected from 28 patients with IIH. Clinical characteristics were assessed, and a genetic association study was performed by sequencing the AQP4 gene on chromosome 18. Genetic data were compared with 52 healthy controls and matched by age, sex and ethnicity. Chi-square test and linear discriminant analysis (LDA) were used in the search of a genotype-phenotype association. RESULTS: While the majority of patients responded to medical treatment, four required shunt application. All, except one, had a good visual outcome. The 24 AQP4 gene SNPs showed no association with IIH. Full cross-validation of the LDA modelling resulted in only 55.1% correct classification of the cases and controls, with a corresponding estimated p-value 0.37. CONCLUSIONS: Our genetic case-control study did not indicate an association between AQP4 gene variants and IIH. However, the theory of an etiopathogenic link between IIH and AQP4 is tempting, and discussed in this article. Association studies with large sample size are difficult to perform owing is the rarity of the condition.

Multi-way metamodelling facilitates insight into the complex input-output maps of nonlinear dynamic models.

BACKGROUND: Statistical approaches to describing the behaviour, including the complex relationships between input parameters and model outputs, of nonlinear dynamic models (referred to as metamodelling) are gaining more and more acceptance as a means for sensitivity analysis and to reduce computational demand. Understanding such input-output maps is necessary for efficient model construction and validation. Multi-way metamodelling provides the opportunity to retain the block-wise structure of the temporal data typically generated by dynamic models throughout the analysis. Furthermore, a cluster-based approach to regional metamodelling allows description of highly nonlinear input-output relationships, revealing additional patterns of covariation. RESULTS: By presenting the N-way Hierarchical Cluster-based Partial Least Squares Regression (N-way HC-PLSR) method, we here combine multi-way analysis with regional cluster-based metamodelling, together making a powerful methodology for extensive exploration of the input-output maps of complex dynamic models. We illustrate the potential of the N-way HC-PLSR by applying it both to predict model outputs as functions of the input parameters, and in the inverse direction (predicting input parameters from the model outputs), to analyse the behaviour of a dynamic model of the mammalian circadian clock. Our results display a more complete cartography of how variation in input parameters is reflected in the temporal behaviour of multiple model outputs than has been previously reported. CONCLUSIONS: Our results indicated that the N-way HC-PLSR metamodelling provides a gain in insight into which parameters that are related to a specific model output behaviour, as well as variations in the model sensitivity to certain input parameters across the model output space. Moreover, the N-way approach allows a more transparent and detailed exploration of the temporal dimension of complex dynamic models, compared to alternative 2-way methods.

Hierarchical cluster-based partial least squares regression (HC-PLSR) is an efficient tool for metamodelling of nonlinear dynamic models.

BACKGROUND: Deterministic dynamic models of complex biological systems contain a large number of parameters and state variables, related through nonlinear differential equations with various types of feedback. A metamodel of such a dynamic model is a statistical approximation model that maps variation in parameters and initial conditions (inputs) to variation in features of the trajectories of the state variables (outputs) throughout the entire biologically relevant input space. A sufficiently accurate mapping can be exploited both instrumentally and epistemically. Multivariate regression methodology is a commonly used approach for emulating dynamic models. However, when the input-output relations are highly nonlinear or non-monotone, a standard linear regression approach is prone to give suboptimal results. We therefore hypothesised that a more accurate mapping can be obtained by locally linear or locally polynomial regression. We present here a new method for local regression modelling, Hierarchical Cluster-based PLS regression (HC-PLSR), where fuzzy C-means clustering is used to separate the data set into parts according to the structure of the response surface. We compare the metamodelling performance of HC-PLSR with polynomial partial least squares regression (PLSR) and ordinary least squares (OLS) regression on various systems: six different gene regulatory network models with various types of feedback, a deterministic mathematical model of the mammalian circadian clock and a model of the mouse ventricular myocyte function. RESULTS: Our results indicate that multivariate regression is well suited for emulating dynamic models in systems biology. The hierarchical approach turned out to be superior to both polynomial PLSR and OLS regression in all three test cases. The advantage, in terms of explained variance and prediction accuracy, was largest in systems with highly nonlinear functional relationships and in systems with positive feedback loops. CONCLUSIONS: HC-PLSR is a promising approach for metamodelling in systems biology, especially for highly nonlinear or non-monotone parameter to phenotype maps. The algorithm can be flexibly adjusted to suit the complexity of the dynamic model behaviour, inviting automation in the metamodelling of complex systems.

Temporal lobe epilepsy and matrix metalloproteinase 9: a tempting relation but negative genetic association.

OBJECTIVE: Neuroplasticity can be defined as the ability of the brain to adapt to environmental impacts. These adaptations include synapse formation and elimination, cortical reorganization, and neurogenesis. In epilepsy these mechanisms may become detrimental and contribute to disease progression. It has been proposed that Matrix Metalloproteinase 9 (MMP-9), a proteinase that cleaves extracellular matrix molecules, may be critically involved in aberrant synaptic formation in hippocampi of patients with Temporal Lobe Epilepsy (TLE). Here we present a case-control study designed to identify possible variants of the MMP-9 gene associated with human TLE. MATERIAL AND METHODS: 218 Norwegian patients with TLE and 181 ethnically matched controls were compared in our association analysis. We also studied associations within two subgroups of TLE--Mesial Temporal Lobe Epilepsy with Hippocampal Sclerosis (MTLE-HS), and Temporal Lobe Epilepsy with childhood Febrile Seizures (TLE-FS). Single nucleotide polymorphisms (SNPs) were selected from HapMap and dbSNP databases for the MMP-9 gene on chromosome 20. We used standard haplotype analysis and multivariate explorative analysis. RESULTS: There were no statistically significant associations between the analyzed SNPs in the MMP-9 gene and TLE, nor were any significant associations found with the two examined subgroups MTLE-HS and TLE-FS, confirmed by both analyses. CONCLUSION: We could not identify any polymorphisms of the human MMP-9 gene that were associated with TLE, MTLE-HS or TLE-FS, in the selected SNPs. However, factors that influence MMP-9 gene expression, post-transcriptional modifications, or the balance between activation and inhibition of MMP-9 may play a role in the pathogenesis of TLE and other epileptic syndromes.

The Response of Enterococcus faecalis V583 to Chloramphenicol Treatment.

Many Enterococcus faecalis strains display tolerance or resistance to many antibiotics, but genes that contribute to the resistance cannot be specified. The multiresistant E. faecalis V583, for which the complete genome sequence is available, survives and grows in media containing relatively high levels of chloramphenicol. No specific genes coding for chloramphenicol resistance has been recognized in V583. We used microarrays to identify genes and mechanisms behind the tolerance to chloramphenicol in V583, by comparison of cells treated with subinhibitory concentrations of chloramphenicol and untreated V583 cells. During a time course experiment, more than 600 genes were significantly differentially transcribed. Since chloramphenicol affects protein synthesis in bacteria, many genes involved in protein synthesis, for example, genes for ribosomal proteins, were induced. Genes involved in amino acid biosynthesis, for example, genes for tRNA synthetases and energy metabolism were downregulated, mainly. Among the upregulated genes were EF1732 and EF1733, which code for potential chloramphenicol transporters. Efflux of drug out of the cells may be one mechanism used by V583 to overcome the effect of chloramphenicol.

Variants of the genes encoding AQP4 and Kir4.1 are associated with subgroups of patients with temporal lobe epilepsy.

OBJECTIVE: The etiopathogenesis of temporal lobe epilepsy (TLE) and its subgroups - mesial temporal lobe epilepsy with hippocampal sclerosis (MTLE-HS) and TLE with antecedent febrile seizures (TLE-FS) - is poorly understood. It has been proposed that the water channel aquaporin-4 (AQP4) and the potassium channel Kir4.1 (KCNJ10 gene) act in concert to regulate extracellular K(+) homeostasis and that functional alterations of these channels influence neuronal excitability. The current study was designed to identify variants of the AQP4 and KCNJ10 genes associated with TLE and subgroups of this condition. MATERIAL AND METHODS: We included 218 Norwegian patients with TLE and 181 ethnically matched healthy controls. An association study was established in which all TLE patients were compared with healthy controls. Additionally, subgroups of 56 MTLE-HS patients were compared with 162 TLE patients without HS, and 102 TLE-FS patients were compared with 105 TLE without FS. RESULTS: We found eight single SNPs, seven in KCNJ10 and one between KCNJ10 and KCNJ9, associated with TLE-FS (nominal p-values from 0.009 to 0.041). Seven of the SNPs segregate into one large haplotype block expanding from KCNJ10 to KCNJ9, including the region interposed those genes. One haplotype was overrepresented in the TLE-FS cases (nominal p-value 0.014). These results were confirmed by explorative multivariate analysis indicating that a combination of SNPs from KCNJ10, the region between KCNJ10 and KCNJ9, and the AQP4 gene is associated with TLE-FS. For the TLE cohort as a whole, explorative multivariate analysis indicated a combination of SNPs from the KCNJ10 and AQP4 genes in association with TLE. CONCLUSION: Variations in the AQP4 and the KCNJ10/KCNJ9 region are likely to be associated with TLE, particularly TLE-FS, supporting the suggestion that perturbations of water and K(+) transport are involved in the etiopathogenesis of TLE.

Estimation of thalamocortical and intracortical network models from joint thalamic single-electrode and cortical laminar-electrode recordings in the rat barrel system.

A new method is presented for extraction of population firing-rate models for both thalamocortical and intracortical signal transfer based on stimulus-evoked data from simultaneous thalamic single-electrode and cortical recordings using linear (laminar) multielectrodes in the rat barrel system. Time-dependent population firing rates for granular (layer 4), supragranular (layer 2/3), and infragranular (layer 5) populations in a barrel column and the thalamic population in the homologous barreloid are extracted from the high-frequency portion (multi-unit activity; MUA) of the recorded extracellular signals. These extracted firing rates are in turn used to identify population firing-rate models formulated as integral equations with exponentially decaying coupling kernels, allowing for straightforward transformation to the more common firing-rate formulation in terms of differential equations. Optimal model structures and model parameters are identified by minimizing the deviation between model firing rates and the experimentally extracted population firing rates. For the thalamocortical transfer, the experimental data favor a model with fast feedforward excitation from thalamus to the layer-4 laminar population combined with a slower inhibitory process due to feedforward and/or recurrent connections and mixed linear-parabolic activation functions. The extracted firing rates of the various cortical laminar populations are found to exhibit strong temporal correlations for the present experimental paradigm, and simple feedforward population firing-rate models combined with linear or mixed linear-parabolic activation function are found to provide excellent fits to the data. The identified thalamocortical and intracortical network models are thus found to be qualitatively very different. While the thalamocortical circuit is optimally stimulated by rapid changes in the thalamic firing rate, the intracortical circuits are low-pass and respond most strongly to slowly varying inputs from the cortical layer-4 population.

Combination of statistical approaches for analysis of 2-DE data gives complementary results.

Five methods for finding significant changes in proteome data have been used to analyze a two-dimensional gel electrophoresis data set. We used both univariate (ANOVA) and multivariate (Partial Least Squares with jackknife, Cross Model Validation, Power-PLS and CovProc) methods. The gels were taken from a time-series experiment exploring the changes in metabolic enzymes in bovine muscle at five time-points after slaughter. The data set consisted of 1377 protein spots, and for each analysis, the data set were preprocessed to fit the requirements of the chosen method. The generated results were one list from each analysis method of proteins found to be significantly changed according to the experimental design. Although the number of selected variables varied between the methods, we found that this was dependent on the specific aim of each method. CovProc and P-PLS focused more on getting the minimum necessary subset of proteins to explain properties of the samples. These methods ended up with less selected proteins. There was also a correlation between level of significance and frequency of selection for the selected proteins.

Pixel-based analysis of multiple images for the identification of changes: a novel approach applied to unravel proteome patterns [corrected] of 2-D electrophoresis gel images.

A novel approach for revealing patterns of proteome variation among series of 2-DE gel images is presented. The approach utilises image alignment to ensure that each pixel represents the same information across all gels. Gel images are normalised, and background corrected, followed by unfolding of the images to 1-D pixel vectors and analysing pixel vectors by multivariate data modelling. Information resulting from the data analysis is refolded back to the image domain for visualisation and interpretation. The method is rapid and suitable for automatic routines applied after the gel alignment. The approach is compared with spot volume analysis to illustrate how this approach can solve persistent problems like mismatch of protein spots, erroneous missing values and failure to detect variation in overlapping proteins. The method may also detect variation in the border area of saturated proteins. The approach is given the name pixel-based analysis of multiple images for the identification of changes (PMC). The method can be used for multiple images in general. Effects of pretreatment of the images are discussed.