On probability-based inference under data missing by design

Whether a statistician wants to complement a probability model for observed data with a prior distribution and carry out fully probabilistic inference, or base the inference only on the likelihood function, may be a fundamental question in theory, but in practice it may well be of less importance if the likelihood contains much more information than the prior. Maximum likelihood inference can be justified as a Gaussian approximation at the posterior mode, using flat priors. However, in situations where parametric assumptions in standard statistical models would be too rigid, more flexible model formulation, combined with fully probabilistic inference, can be achieved using hierarchical Bayesian parametrization. This work includes five articles, all of which apply probability modeling under various problems involving incomplete observation. Three of the papers apply maximum likelihood estimation and two of them hierarchical Bayesian modeling. Because maximum likelihood may be presented as a special case of Bayesian inference, but not the other way round, in the introductory part of this work we present a framework for probability-based inference using only Bayesian concepts. We also re-derive some results presented in the original articles using the toolbox equipped herein, to show that they are also justifiable under this more general framework. Here the assumption of exchangeability and de Finetti's representation theorem are applied repeatedly for justifying the use of standard parametric probability models with conditionally independent likelihood contributions. It is argued that this same reasoning can be applied also under sampling from a finite population. The main emphasis here is in probability-based inference under incomplete observation due to study design. This is illustrated using a generic two-phase cohort sampling design as an example. The alternative approaches presented for analysis of such a design are full likelihood, which utilizes all observed information, and conditional likelihood, which is restricted to a completely observed set, conditioning on the rule that generated that set. Conditional likelihood inference is also applied for a joint analysis of prevalence and incidence data, a situation subject to both left censoring and left truncation. Other topics covered are model uncertainty and causal inference using posterior predictive distributions. We formulate a non-parametric monotonic regression model for one or more covariates and a Bayesian estimation procedure, and apply the model in the context of optimal sequential treatment regimes, demonstrating that inference based on posterior predictive distributions is feasible also in this case.

Discovering hidden structures in molecular data using a Bayesian partition model approach

Advancements in the analysis techniques have led to a rapid accumulation of biological data in databases. Such data often are in the form of sequences of observations, examples including DNA sequences and amino acid sequences of proteins. The scale and quality of the data give promises of answering various biologically relevant questions in more detail than what has been possible before. For example, one may wish to identify areas in an amino acid sequence, which are important for the function of the corresponding protein, or investigate how characteristics on the level of DNA sequence affect the adaptation of a bacterial species to its environment. Many of the interesting questions are intimately associated with the understanding of the evolutionary relationships among the items under consideration. The aim of this work is to develop novel statistical models and computational techniques to meet with the challenge of deriving meaning from the increasing amounts of data. Our main concern is on modeling the evolutionary relationships based on the observed molecular data. We operate within a Bayesian statistical framework, which allows a probabilistic quantification of the uncertainties related to a particular solution. As the basis of our modeling approach we utilize a partition model, which is used to describe the structure of data by appropriately dividing the data items into clusters of related items. Generalizations and modifications of the partition model are developed and applied to various problems. Large-scale data sets provide also a computational challenge. The models used to describe the data must be realistic enough to capture the essential features of the current modeling task but, at the same time, simple enough to make it possible to carry out the inference in practice. The partition model fulfills these two requirements. The problem-specific features can be taken into account by modifying the prior probability distributions of the model parameters. The computational efficiency stems from the ability to integrate out the parameters of the partition model analytically, which enables the use of efficient stochastic search algorithms.

Causes and consequences of variation in nestling immune function

Defence against pathogens is a vital need of all living organisms that has led to the evolution of complex immune mechanisms. However, although immunocompetence the ability to resist pathogens and control infection has in recent decades become a focus for research in evolutionary ecology, the variation in immune function observed in natural populations is relatively little understood. This thesis examines sources of this variation (environmental, genetic and maternal effects) during the nestling stage and its fitness consequences in wild populations of passerines: the blue tit (Cyanistes caeruleus) and the collared flycatcher (Ficedula albicollis). A developing organism may face a dilemma as to whether to allocate limited resources to growth or to immune defences. The optimal level of investment in immunity is shaped inherently by specific requirements of the environment. If the probability of contracting infection is low, maintaining high growth rates even at the expense of immune function may be advantageous for nestlings, as body mass is usually a good predictor of post-fledging survival. In experiments with blue tits and haematophagous hen fleas (Ceratophyllus gallinae) using two methods, methionine supplementation (to manipulate nestlings resource allocation to cellular immune function) and food supplementation (to increase resource availability), I confirmed that there is a trade-off between growth and immunity and that the abundance of ectoparasites is an environmental factor affecting allocation of resources to immune function. A cross-fostering experiment also revealed that environmental heterogeneity in terms of abundance of ectoparasites may contribute to maintaining additive genetic variation in immunity and other traits. Animal model analysis of extensive data collected from the population of collared flycatchers on Gotland (Sweden) allowed examination of the narrow-sense heritability of PHA-response the most commonly used index of cellular immunocompetence in avian studies. PHA-response is not heritable in this population, but is subject to a non-heritable origin (presumably maternal) effect. However, experimental manipulation of yolk androgen levels indicates that the mechanism of the maternal effect in PHA-response is not in ovo deposition of androgens. The relationship between PHA-response and recruitment was studied for over 1300 collared flycatcher nestlings. Multivariate selection analysis shows that it is body mass, not PHA-response, that is under direct selection. PHA-response appears to be related to recruitment because of its positive relationship with body mass. These results imply that either PHA-response fails to capture the immune mechanisms that are relevant for defence against pathogens encountered by fledglings or that the selection pressure from parasites is not as strong as commonly assumed.

Finnish Dosimetric Practice for Epithermal Neutron Beam Dosimetry in Boron Neutron Capture Therapy

Boron neutron capture therapy (BNCT) is a form of chemically targeted radiotherapy that utilises the high neutron capture cross-section of boron-10 isotope to achieve a preferential dose increase in the tumour. The BNCT dosimetry poses a special challenge as the radiation dose absorbed by the irradiated tissues consists of several dose different components. Dosimetry is important as the effect of the radiation on the tissue is correlated with the radiation dose. Consistent and reliable radiation dose delivery and dosimetry are thus basic requirements for radiotherapy. The international recommendations for are not directly applicable to BNCT dosimetry. The existing dosimetry guidance for BNCT provides recommendations but also calls for investigating for complementary methods for comparison and improved accuracy. In this thesis the quality assurance and stability measurements of the neutron beam monitors used in dose delivery are presented. The beam monitors were found not to be affected by the presence of a phantom in the beam and that the effect of the reactor core power distribution was less than 1%. The weekly stability test with activation detectors has been generally reproducible within the recommended tolerance value of 2%. An established toolkit for epithermal neutron beams for determination of the dose components is presented and applied in an international dosimetric intercomparison. The measured quantities (neutron flux, fast neutron and photon dose) by the groups in the intercomparison were generally in agreement within the stated uncertainties. However, the uncertainties were large, ranging from 3-30% (1 standard deviation), emphasising the importance of dosimetric intercomparisons if clinical data is to be compared between different centers. Measurements with the Exradin type 2M ionisation chamber have been repeated in the epithermal neutron beam in the same measurement configuration over the course of 10 years. The presented results exclude severe sensitivity changes to thermal neutrons that have been reported for this type of chamber. Microdosimetry and polymer gel dosimetry as complementary methods for epithermal neutron beam dosimetry are studied. For microdosimetry the comparison of results with ionisation chambers and computer simulation showed that the photon dose measured with microdosimetry was lower than with the two other methods. The disagreement was within the uncertainties. For neutron dose the simulation and microdosimetry results agreed within 10% while the ionisation chamber technique gave 10-30% lower neutron dose rates than the two other methods. The response of the BANG-3 gel was found to be linear for both photon and epithermal neutron beam irradiation. The dose distribution normalised to dose maximum measured by MAGIC polymer gel was found to agree well with the simulated result near the dose maximum while the spatial difference between measured and simulated 30% isodose line was more than 1 cm. In both the BANG-3 and MAGIC gel studies, the interpretation of the results was complicated by the presence of high-LET radiation.

Proton magnetic resonance spectroscopy of a boron neutron capture therapy 10B-carrier, L-p-boronophenylalanine-fructose complex

Boron neutron capture therapy (BNCT) is a radiotherapy that has mainly been used to treat malignant brain tumours, melanomas, and head and neck cancer. In BNCT, the patient receives an intravenous infusion of a 10B-carrier, which accumulates in the tumour area. The tumour is irradiated with epithermal or thermal neutrons, which result in a boron neutron capture reaction that generates heavy particles to damage tumour cells. In Finland, boronophenylalanine fructose (BPA-F) is used as the 10B-carrier. Currently, the drifting of boron from blood to tumour as well as the spatial and temporal accumulation of boron in the brain, are not precisely known. Proton magnetic resonance spectroscopy (1H MRS) could be used for selective BPA-F detection and quantification as aromatic protons of BPA resonate in the spectrum region, which is clear of brain metabolite signals. This study, which included both phantom and in vivo studies, examined the validity of 1H MRS as a tool for BPA detection. In the phantom study, BPA quantification was studied at 1.5 and 3.0 T with single voxel 1H MRS, and at 1.5 T with magnetic resonance imaging (MRSI). The detection limit of BPA was determined in phantom conditions at 1.5 T and 3.0 T using single voxel 1H MRS, and at 1.5 T using MRSI. In phantom conditions, BPA quantification accuracy of ± 5% and ± 15% were achieved with single voxel MRS using external or internal (internal water signal) concentration references, respectively. For MRSI, a quantification accuracy of <5% was obtained using an internal concentration reference (creatine). The detection limits of BPA in phantom conditions for the PRESS sequence were 0.7 (3.0 T) and 1.4 mM (1.5 T) mM with 20 × 20 × 20 mm3 single voxel MRS, and 1.0 mM with acquisition-weighted MRSI (nominal voxel volume 10(RL) × 10(AP) × 7.5(SI) mm3), respectively. In the in vivo study, an MRSI or single voxel MRS or both was performed for ten patients (patients 1-10) on the day of BNCT. Three patients had glioblastoma multiforme (GBM), and five patients had a recurrent or progressing GBM or anaplastic astrocytoma gradus III, and two patients had head and neck cancer. For nine patients (patients 1-9), MRS/MRSI was performed 70-140 min after the second irradiation field, and for one patient (patient 10), the MRSI study began 11 min before the end of the BPA-F infusion and ended 6 min after the end of the infusion. In comparison, single voxel MRS was performed before BNCT, for two patients (patients 3 and 9), and for one patient (patient 9), MRSI was performed one month after treatment. For one patient (patient 10), MRSI was performed four days before infusion. Signals from the tumour spectrum aromatic region were detected on the day of BNCT in three patients, indicating that in favourable cases, it is possible to detect BPA in vivo in the patient’s brain after BNCT treatment or at the end of BPA-F infusion. However, because the shape and position of the detected signals did not exactly match the BPA spectrum detected in the in vitro conditions, assignment of BPA is difficult. The opportunity to perform MRS immediately after the end of BPA-F infusion for more patients is necessary to evaluate the suitability of 1H MRS for BPA detection or quantification for treatment planning purposes. However, it could be possible to use MRSI as criteria in selecting patients for BNCT.

Explorations in the distributional and semantic similarity of words

A straightforward computation of the list of the words (the `tail words' of the list) that are distributionally most similar to a given word (the `head word' of the list) leads to the question: How semantically similar to the head word are the tail words; that is: how similar are their meanings to its meaning? And can we do better? The experiment was done on nearly 18,000 most frequent nouns in a Finnish newsgroup corpus. These nouns are considered to be distributionally similar to the extent that they occur in the same direct dependency relations with the same nouns, adjectives and verbs. The extent of the similarity of their computational representations is quantified with the information radius. The semantic classification of head-tail pairs is intuitive; some tail words seem to be semantically similar to the head word, some do not. Each such pair is also associated with a number of further distributional variables. Individually, their overlap for the semantic classes is large, but the trained classification-tree models have some success in using combinations to predict the semantic class. The training data consists of a random sample of 400 head-tail pairs with the tail word ranked among the 20 distributionally most similar to the head word, excluding names. The models are then tested on a random sample of another 100 such pairs. The best success rates range from 70% to 92% of the test pairs, where a success means that the model predicted my intuitive semantic class of the pair. This seems somewhat promising when distributional similarity is used to capture semantically similar words. This analysis also includes a general discussion of several different similarity formulas, arranged in three groups: those that apply to sets with graded membership, those that apply to the members of a vector space, and those that apply to probability mass functions.