Data fusion and matching by maximizing statistical dependencies

The core aim of machine learning is to make a computer program learn from the experience. Learning from data is usually defined as a task of learning regularities or patterns in data in order to extract useful information, or to learn the underlying concept. An important sub-field of machine learning is called multi-view learning where the task is to learn from multiple data sets or views describing the same underlying concept. A typical example of such scenario would be to study a biological concept using several biological measurements like gene expression, protein expression and metabolic profiles, or to classify web pages based on their content and the contents of their hyperlinks. In this thesis, novel problem formulations and methods for multi-view learning are presented. The contributions include a linear data fusion approach during exploratory data analysis, a new measure to evaluate different kinds of representations for textual data, and an extension of multi-view learning for novel scenarios where the correspondence of samples in the different views or data sets is not known in advance. In order to infer the one-to-one correspondence of samples between two views, a novel concept of multi-view matching is proposed. The matching algorithm is completely data-driven and is demonstrated in several applications such as matching of metabolites between humans and mice, and matching of sentences between documents in two languages.

Genetic variation in the LDL receptor-related protein 5 (LRP5) gene : Association with bone health and metabolic parameters

Bone mass accrual and maintenance are regulated by a complex interplay between genetic and environmental factors. Recent studies have revealed an important role for the low-density lipoprotein receptor-related protein 5 (LRP5) in this process. The aim of this thesis study was to identify novel variants in the LRP5 gene and to further elucidate the association of LRP5 and its variants with various bone health related clinical characteristics. The results of our studies show that loss-of-function mutations in LRP5 cause severe osteoporosis not only in homozygous subjects but also in the carriers of these mutations, who have significantly reduced bone mineral density (BMD) and increased susceptibility to fractures. In addition, we demonstrated for the first time that a common polymorphic LRP5 variant (p.A1330V) was associated with reduced peak bone mass, an important determinant of BMD and osteoporosis in later life. The results from these two studies are concordant with results seen in other studies on LRP5 mutations and in association studies linking genetic variation in LRP5 with BMD and osteoporosis. Several rare LRP5 variants were identified in children with recurrent fractures. Sequencing and multiplex ligation-dependent probe amplification (MLPA) analyses revealed no disease-causing mutations or whole-exon deletions. Our findings from clinical assessments and family-based genotype-phenotype studies suggested that the rare LRP5 variants identified are not the definite cause of fractures in these children. Clinical assessments of our study subjects with LPR5 mutations revealed an unexpectedly high prevalence of impaired glucose tolerance and dyslipidaemia. Moreover, in subsequent studies we discovered that common polymorphic LRP5 variants are associated with unfavorable metabolic characteristics. Changes in lipid profile were already apparent in pre-pubertal children. These results, together with the findings from other studies, suggest an important role for LRP5 also in glucose and lipid metabolism. Our results underscore the important role of LRP5 not only in bone mass accrual and maintenance of skeletal health but also in glucose and lipid metabolism. The role of LRP5 in bone metabolism has long been studied, but further studies with larger study cohorts are still needed to evaluate the specific role of LRP5 variants as metabolic risk factors.

Compliance of rough cylinders in compression

Compression of a rough turned cylinder between two hard, smooth, flat plates has been analysed with the aid of a mathematical model based on statistical analysis. It is assumed that the asperity peak heights follow Gaussian or normal and beta distribution functions and that the loaded asperities comply as though they are completely isolated from the neighbouring ones. Equations have been developed for the loadcompliance relation of the real surface using a simplified relation of the form W0 = K1δn for the load-compliance of a single asperity. Parameters K1 and n have considerable influence on the load-compliance curve and they depend on the material, tip angle of the asperity, standard deviation of the asperity peak height distribution and the density of the asperities.

Moment-Matched Lognormal Modeling of Uplink Interference with Power Control and Cell Selection

We develop an alternate characterization of the statistical distribution of the inter-cell interference power observed in the uplink of CDMA systems. We show that the lognormal distribution better matches the cumulative distribution and complementary cumulative distribution functions of the uplink interference than the conventionally assumed Gaussian distribution and variants based on it. This is in spite of the fact that many users together contribute to uplink interference, with the number of users and their locations both being random. Our observations hold even in the presence of power control and cell selection, which have hitherto been used to justify the Gaussian distribution approximation. The parameters of the lognormal are obtained by matching moments, for which detailed analytical expressions that incorporate wireless propagation, cellular layout, power control, and cell selection parameters are developed. The moment-matched lognormal model, while not perfect, is an order of magnitude better in modeling the interference power distribution.

ErbB-2 expression and its association with other biological parameters of breast cancer among Indian women

Overexpression of the epidermal growth factor receptor family genes, which include ErbB-1, 2, 3 and 4, has been implicated in a number of cancers. We have studied the extent of ErbB-2 overexpression among Indian women with sporadic breast cancer. Methods: Immmunohistochemistry and genomic polymerase chain reaction (PCR) were used to study the ErbB2 overexpression. ErbB2 status was correlated with other clinico-pathological parameters, including patient survival. Results: ErbB-2 overexpression was detected in 43.2% (159/368) of the cases by immunohistochemistry. For a sub-set of patients (n = 55) for whom total DNA was available, ErbB-2 gene amplification was detected in 25.5% (14/55) of the cases by genomic PCR. While the ErbB2 overexpression was significantly higher in patients with lymphnode (χ2 = 12.06, P≤ 0.001), larger tumor size (χ2 = 8.22, P = 0.042) and ductal carcinoma (χ2 = 15.42, P ≤ 0.001), it was lower in patients with disease-free survival (χ2 = 22.13, P ≤ 0.001). Survival analysis on a sub-set of patients for whom survival data were available (n = 179) revealed that ErbB-2 status (χ2 =25.94, P ≤ 0.001), lymphnode status (χ2 = 12.68, P ≤ 0.001), distant metastasis (χ2 = 19.49, P ≤ 0.001) and stage of the disease (χ2 = 28.04, P ≤0.001) were markers of poor prognosis. Conclusions: ErbB-2 overexpression was significantly greater compared with the Western literature, but comparable to other Indian studies. Significant correlation was found between ErbB-2 status and lymphnode status, tumor size and ductal carcinoma. ErbB-2 status, lymph node status, distant metastasis and stage of the disease were found to be prognostic indicators.

Prediction of swelling pressure of soil using artificial intelligence techniques

The swelling pressure of soil depends upon various soil parameters such as mineralogy, clay content, Atterberg's limits, dry density, moisture content, initial degree of saturation, etc. along with structural and environmental factors. It is very difficult to model and analyze swelling pressure effectively taking all the above aspects into consideration. Various statistical/empirical methods have been attempted to predict the swelling pressure based on index properties of soil. In this paper, the computational intelligence techniques artificial neural network and support vector machine have been used to develop models based on the set of available experimental results to predict swelling pressure from the inputs; natural moisture content, dry density, liquid limit, plasticity index, and clay fraction. The generalization of the model to new set of data other than the training set of data is discussed which is required for successful application of a model. A detailed study of the relative performance of the computational intelligence techniques has been carried out based on different statistical performance criteria.

Prediction of swelling pressure of soil using artificial intelligence techniques

The swelling pressure of soil depends upon various soil parameters such as mineralogy, clay content, Atterberg's limits, dry density, moisture content, initial degree of saturation, etc. along with structural and environmental factors. It is very difficult to model and analyze swelling pressure effectively taking all the above aspects into consideration. Various statistical/empirical methods have been attempted to predict the swelling pressure based on index properties of soil. In this paper, the computational intelligence techniques artificial neural network and support vector machine have been used to develop models based on the set of available experimental results to predict swelling pressure from the inputs; natural moisture content, dry density, liquid limit, plasticity index, and clay fraction. The generalization of the model to new set of data other than the training set of data is discussed which is required for successful application of a model. A detailed study of the relative performance of the computational intelligence techniques has been carried out based on different statistical performance criteria.

Penetration of hydromagnetic energy deep into the magnetosphere

Magnetometer data, acquired on spacecraft and simultaneously at high and low latitudes on the ground, are compared in order to study the propagation characteristics of hydromagnetic energy deep into the magnetosphere. Single events provide evidence that wave energy at L ∼ 3 can at times be only one order of magnitude lower than at L ∼ 13. In addition, statistical analyses of the H-component groundbased data obtained during local daytime hours of 17 July-3 August 1985 show that wave amplitudes at L ∼ 3 are generally 10-30 times lower than at L ∼ 13. The L-dependence of near-equator magnetic field fluctuations measured on ISEE-2 show a sharp drop in energy near the magnetopause and a more gradual fall-off of energy deeper inside the magnetosphere. Such high levels of wave power deep in the magnetosphere have not been quantitatively understood previously. Our initial attempt is to calculate the decay length of an evanescent wave generated at a thick magnetopause boundary. Numerical calculations show that fast magnetosonic modes (called magnetopause and inner mode) can be generated under very restrictive conditions for the field and plasma parameters. These fast compressional modes may have their energy reduced by only one order of magnitude over a penetration depth of about 8RE. More realistic numerical simulations need to be carried out to see whether better agreement with the data can be attained.

Estimation of signals in colored non gaussian noise based on Gaussian Mixture models

Non-Gaussianity of signals/noise often results in significant performance degradation for systems, which are designed using the Gaussian assumption. So non-Gaussian signals/noise require a different modelling and processing approach. In this paper, we discuss a new Bayesian estimation technique for non-Gaussian signals corrupted by colored non Gaussian noise. The method is based on using zero mean finite Gaussian Mixture Models (GMMs) for signal and noise. The estimation is done using an adaptive non-causal nonlinear filtering technique. The method involves deriving an estimator in terms of the GMM parameters, which are in turn estimated using the EM algorithm. The proposed filter is of finite length and offers computational feasibility. The simulations show that the proposed method gives a significant improvement compared to the linear filter for a wide variety of noise conditions, including impulsive noise. We also claim that the estimation of signal using the correlation with past and future samples leads to reduced mean squared error as compared to signal estimation based on past samples only.

Simulating and Forecasting the Demand for New Consumer Durables

A diffusion/replacement model for new consumer durables designed to be used as a long-term forecasting tool is developed. The model simulates new demand as well as replacement demand over time. The model is called DEMSIM and is built upon a counteractive adoption model specifying the basic forces affecting the adoption behaviour of individual consumers. These forces are the promoting forces and the resisting forces. The promoting forces are further divided into internal and external influences. These influences are operationalized within a multi-segmental diffusion model generating the adoption behaviour of the consumers in each segment as an expected value. This diffusion model is combined with a replacement model built upon the same segmental structure as the diffusion model. This model generates, in turn, the expected replacement behaviour in each segment. To be able to use DEMSIM as a forecasting tool in early stages of a diffusion process estimates of the model parameters are needed as soon as possible after product launch. However, traditional statistical techniques are not very helpful in estimating such parameters in early stages of a diffusion process. To enable early parameter calibration an optimization algorithm is developed by which the main parameters of the diffusion model can be estimated on the basis of very few sales observations. The optimization is carried out in iterative simulation runs. Empirical validations using the optimization algorithm reveal that the diffusion model performs well in early long-term sales forecasts, especially as it comes to the timing of future sales peaks.

Apsidal motion in 4U 0115+63 and orbital parameters of 2S 1417-624 and V0332+53

Be/X-ray binary pulsars have wide eccentric orbits and hence the angle of periastron of the orbit is very well defined in these sources. The presence of an X-ray pulsar allows for accurate measurements of orbital elements. A Be star usually is a rapidly rotating star and hence will deviate from spherical geometry. The tidal interaction between the neutron star and the Be star will add to the distortion of the Be star and alter its mass distribution. Thus a measurable rate of apsidal motion is expected from these systems. In this paper, we present the first conclusive detection of apsidal motion of the binary 4U 0115+63. We also present new and accurate orbital parameters of the Be/X-ray binaries V0332+53 and 2S 1417-624.

Electrocardiographic parameters of ventricular repolarization : modifiers and the prognostic value

Electric activity of the heart consists of repeated cardiomyocyte depolarizations and repolarizations. Abnormalities in repolarization predispose to ventricular arrhythmias. In body surface electrocardiogram, ventricular repolarization generates the T wave. Several electrocardiographic measures have been developed both for clinical and research purposes to detect repolarization abnormalities. The study aim was to investigate modifiers of ventricular repolarization with the focus on the relationship of the left ventricular mass, antihypertensive drugs, and common gene variants, to electrocardiographic repolarization parameters. The prognostic value of repolarization parameters was also assessed. The study subjects originated from a population of more than 200 middle-aged hypertensive men attending the GENRES hypertension study, and from an epidemiological survey, the Health 2000 Study, including more than 6000 participants. Ventricular repolarization was analysed from digital standard 12-lead resting electrocardiograms with two QT-interval based repolarization parameters (QT interval, T-wave peak to T-wave end interval) and with a set of four T-wave morphology parameters. The results showed that in hypertensive men, a linear change in repolarization parameters is present even in the normal range of left ventricular mass, and that even mild left ventricular hypertrophy is associated with potentially adverse electrocardiographic repolarization changes. In addition, treatments with losartan, bisoprolol, amlodipine, and hydrochlorothiazide have divergent short-term effects on repolarization parameters in hypertensive men. Analyses of the general population sample showed that single nucleotide polymorphisms in KCNH2, KCNE1, and NOS1AP genes are associated with changes in QT-interval based repolarization parameters but not consistently with T-wave morphology parameters. T-wave morphology parameters, but not QT interval or T-wave peak to T-wave end interval, provided independent prognostic information on mortality. The prognostic value was specifically related to cardiovascular mortality. The results indicate that, in hypertension, altered ventricular repolarization is already present in mild left ventricular mass increase, and that commonly used antihypertensive drugs may relatively rapidly and treatment-specifically modify electrocardiographic repolarization parameters. Common variants in cardiac ion channel genes and NOS1AP gene may also modify repolarization-related arrhythmia vulnerability. In the general population, T-wave morphology parameters may be useful in the risk assessment of cardiovascular mortality.

Shell-model calculation of defect energies in alkali halides employing crystal-independent interionic potential parameters

Shell model calculation of defect energies in alkali halides have been carried out using the ion-dependent, crystal-independent potential parameters of Sangster and Atwood (1978). Results indicate that appreciable differences exist between barrier heights for migration of cations and anions. While barrier heights for cations are generally lower than for anions in alkali halides of NaCl structure, the opposite is true in alkali halides of CsCl structure.

Theoretical Treatment of Interaction Parameters in Multicomponent Metallic Solutions

A new theoretical equation for interaction parameter in multicomponent metallic solutions is developed using the pseudopotential formalism coupled with the free energy of the hard sphere system. The approximate expression for the pseudopotential term is given in terms of the heat of solution at infinite dilution, to allow easy evaluation of the interaction parameter in various multicomponent systems. This theory has been applied to 23 non-ferrous alloys based on Pb, Sn, Bi and indium. Comparison with the results of previous theoretical calculations using only the hard sphere model suggests that the inclusion of the pseudopotential term yields a quantitatively more correct prediction of interaction parameters in multicomponent metallic solutions. Numerical calculations were also made for 320 Fe-base solutions relevant to steelmaking and the agreement between calculation and experimental data appears reasonable, with 90% reliability in predicting the correct sign.