Browsing by study line "Biostatistics and Bioinformatics"

In inductive inference phenomena from the past are modeled in order to make predictions of the future. The mathematical concept of exchangeability for random sequences provides a mathematical justification for the assumption that observations are independently and identically distributed given some underlying parameters estimable from the empirical distribution of the observations. The theory of exchangeability contains basic elements for inductive inference, such as the de Finetti representation theorem for the probability of a general exchangeable sequence, prior probability distributions for the parameters in the representation theorem, as well as the predictive probabilities, or rule of succession, for new observations from the random sequence under consideration. However, entirely unanticipated observations pose a problem for inductive inference. How can one assign a probability for an event that has never been seen before? This is called the sampling of species problem. Under exchangeability, the number of possible different events t has to be known before-hand to be able to assign an equal prior probability 1/t for each event. In the sampling of species problem an assumption of infinite possible events has to be made, leading to the prior probability 1/∞ for each event, which is impossible. Exchangeability is thus inadequate to handle probability distributions for infinite possible events. It turns out that a solution to the sampling of species problem arises from partition exchangeability. Exchangeable random sequences have the same probability of occurring, if the observations in the sequence have identical frequencies. Under partition exchangeability, the sequences have the same probability of occurring when they share identical frequencies of frequencies. In this thesis, partition exchangeability is introduced as a framework of inductive inference by juxtaposing it with the more familiar type of exchangeability for random sequences. Partition exchangeability has parallel elements to exchangeability, in the Kingman representation theorem, the Poisson-Dirichlet distribution for the prior probability distribution, and a corresponding rule of succession. The rules of succession are required in the problem of supervised classification to provide product predictive probabilities to be maximized by assigning the test data into pre-defined classes based on training data. A Bayesian construction of supervised classification is discussed in this thesis. In theory, the best classification performance is gained when assigning the class labels to the test data simultaneously, but because of computational complexity, an assumption is often made where the test data points are i.i.d. with regards to each other. In the case of a known set of possible events these simultaneous and marginal classifiers converge in their test data predictive probabilities as the amount of training data tends to infinity, justifying the use of the simpler marginal classifier with enough training data. These two classifiers are implemented in this thesis under partition exchangeability, and it is shown in theory and in practice with a simulation study that the same asymptotic convergence between the simultaneous and marginal classifiers applies with partition exchangeable data as well. Finally, a small application in single cell RNA expression is explored.

Background/Objectives: Various studies have shown the advantage when incorporating polygenic risk scores (PRSs) in models with classic risk factors. However, systematic comparisons of PRSs with non-genetic factors are lacking. In particular, many studies on PRSs do not even report the predictive performance of the confounders, such as age and sex, included in the model, which are already very predictive for most diseases. We looked at the ability of PRSs to predict the onset of 18 diseases in FinnGen R8 (N=342,499) and compared PRSs with the known non-genetic risk factors, age, sex, Education, and Charlson Comorbidity Index (CCI). Methods: We set up individual studies for the 18 diseases. A single study consisted of an exposure (1999-2009), a washout (2009-2011), and an observation period (2011-2019). Eligible individuals could not have the selected disease of interest inside the disease-free period, which ranged from birth until the beginning of the observation period. We then defined the case and control status based on the diagnoses in the observation period and calculated the phenotypic scores during the exposure period. The PRSs were calculated using MegaPRS and the latest publicly available genome-wide association study summary statistics. We then fitted separate Cox proportional hazards models for each disease to predict disease onset during the observation period. Results: In FinnGen, the model’s predictive ability (c-index) with all predictors ranged from 0.565 (95%CI: 0.552-0.576) for Acute Appendicitis to 0.838 (95% CI: 0.834-0.841) for Atrial Fibrillation. The PRSs outperformed the phenotypic predictors, CCI, and Education, for 6/18 diseases and still significantly enhance onset prediction for 13/18 diseases when added to a model with only non-genetic predictors. Conclusion: Overall, we showed that for many diseases PRSs add predictive power over commonly used predictors - such as age, sex, CCI, and Education. However, many important challenges must be addressed before implementing PRSs in clinical practice. Notably, we will need disease-specific cost- benefit analyses and studies to assess the direct impact of including PRSs in clinical use. Nonetheless, as more research is being conducted, PRSs could play an increasingly valuable role in identifying individuals at higher risk for certain diseases and enabling targeted interventions to improve health outcomes.

Species factories are defined as times and places in the fossil record where and when an exceptionally large number of new species occurs. While several tailored solutions for the mammalian record have been proposed, how to identify species factories computationally in a standardized way is still an open question. To quantify what is exceptional, we first need to quantify what is regular. One of the main challenges in this identification process is to account for sampling unevenness, which depends on several methodological decisions, including the scale of the analysis (aggrega- tion radius). In this thesis we used Capture-Mark-Recapture methods (CMR) with spatial aggregation guided by network modelling, to estimate the sampling probabilities for the species in the NOW database of mammalian fossil occurrences. Since the mammalian record is sparse and most localities include only a few species, we coupled CMR with tailored spatial aggregation approaches to estimate the sampling prob- abilities. We then used these sampling probabilities to quantify background speciation rates and assess what rates are abnormal. We represented aggregated fossil data as a bipartite network and used community detection to evaluate how the choice of an aggre- gation radius impacts the modular structure. After aggregating the data according to the radius chosen using networks analysis, we es- timated sampling probabilities using CMR. These probabilities allow the adjustment for sampling unevenness so that the difference in findings can be compared across locations and cannot be due to differences in sampling. We identified as species factories the locations with origination rate in the highest 5% after adjustment per time unit. Once the species factories had been identified, we looked for paleoecological patterns in these places that may be lacking elsewhere, finding that species factories present a lower number of findings and of different species among findings, but a higher ratio between the amount of different species and of total findings than the rest of the locations. This would indicate that, even if species factories might accommodate fewer species, they present a higher diversity. To make sure these results were not only due to chance, we performed the same analysis on 100 randomized experiments obtained using a modified version of the Curveball Algo- rithm and compared the values obtained from the original dataset and the ones obtained from the randomized ones. This comparison showed us that species factories tend to have more extreme values than the ones obtained through randomization, which would indicate that species factories present specific paleoecological patterns that are not present in other locations.

Sex differences can be found in most human phenotypes, and they play an important role in human health and disease. Females and males have different sex chromosomes, which are known to cause sex differences, as are differences in the concentration of sex hormones such as testosterone, estradiol and progesterone. However, the role of the autosomes has remained more debated. The primary aim of this thesis is to assess the magnitude and relevance of human sex-specific genetic architecture in the autosomes. This is done by calculating sex-specific heritability estimates and genetic correlation estimates between females and males, as well as comparing these to sex differences on the phenotype level. Additionally, the heritability and genetic correlation estimates are compared between two populations, in order to assess the magnitude of sex differences compared to differences between populations. The analyses in this thesis are based on sex-stratified genome-wide association study (GWAS) data from 48 phenotypes in the UK Biobank (UKB), which contains genotype data from approximately 500 000 individuals as well as thousands of phenotype measurements. A replication of the analyses using three phenotypes was also made on data from the FinnGen project, with a dataset from approximately 175 000 individuals. The 48 phenotypes used in this study range from biomarkers such as serum testosterone and albumin levels to general traits such as height and blood pressure. The heritability and genetic correlation estimates were calculated using linkage disequilibrium score regression (LDSC). LDSC fits a linear regression model between test statistic values of GWAS variants and linkage disequilibrium (LD) scores calculated from a reference population. For most phenotypes, the heritability and genetic correlation results show little evidence of sex differences. Serum testosterone level and waist-to-hip ratio are exceptions to this, showing strong evidence of sex differences both on the genetic and the phenotype level. However, the overall correlation between phenotype level sex differences and sex differences in heritability or genetic correlation estimates is low. The replication in the FinnGen dataset for height, weight and body mass index (BMI), showed that for these traits the differences in heritability estimates and genetic correlations between the Finnish and UK populations are comparable or larger than the differences found between males and females.

The aim of this thesis is to predict total career racing performance of Finnish trotter horses by using trotters early career racing performance and other early career variables. This thesis presents a brief introductory of harness racing and horses used in Finnish trotting sport. The data is presented and modified for predictions, with descriptive statistics of tables and visuals. The machine learning method of Random forests for regression is introduced and used in the predictions. After training the model, this thesis presents the prediction accuracy and variables of importance of the predictions of total career racing performance for both Finnhorse trotters and Finnish Standardbred trotter population. Finally, the writer discusses on the shortages and possible improvements for future research. The data for this thesis was provided by The Finnish trotting and breeding association (Suomen Hippos ry), which included all information of harness races from 1984 to the end of 2019, raced in Finland. From almost three million rows, the data was summarised to a data table of 46704 rows of trotters, that have started their career at earliest allowed three age groups. A total of 37 independent variables were used to predict three outcomes of total career earnings, total number of career starts and total number of career first placings, as separate models. The predictors are derived from other studies that estimate the environmental and genetic factors of racing performance of a trotter. The three models performed poor to moderate, with total earnings having the highest prediction accuracy. The model predicted quite well larger amounts of earnings, but was avid to predict some earnings when there in fact were none. Prediction accuracy of total number of starts was poor, especially when the true amount of starts was low. Model that predicted total number of career first placings performed the worst. This can partially be explained by the fact that winning is a rare event for a trotter in general. The models fit better for Finnish Standardbred trotters than for Finnhorse trotters. This thesis works as a good basis for future similar research, where massive amounts of data and machine learning is used to predict trotter’s career, racing performance or other factors. The results show that predicting total career racing performance as a classification problem could be a better fit than regression. These adequate classes, as well as possible better predictors and suitable imputes for missing values, should be consulted with an audience of superior knowledge in harness racing.