Browsing by master's degree program "Life Science Informatics -maisteriohjelma"

Cancer consists of heterogeneous cell populations that repeatedly undergo natural selection. These cell populations contest with each other for space and nutrients and try to generate phenotypes that maximize their ecological fitness. For achieving this, they evolve evolutionarily stable strategies. When an oncologist starts to treat cancer, another game emerges. While affected by the cellular evolution processes, modeling of this game owes to the results of the classical game theory. This thesis investigates the theoretical foundations of adaptive cancer treatment. It draws from two game theoretical approaches, evolutionary game theory and Stackelberg leader-follower game. The underlying hypothesis of adaptive regimen is that the patient's cancer burden can be administered by leveraging the resource competition between treatment-sensitive and treatment-resistant cells. The intercellular competition is mathematically modelled as an evolutionary game using the G function approach. The properties of the evolutionary stability, such as ESS, the ESS maximum principle, and convergence stability, that are relevant to tumorigenesis and intra-tumoral dynamics, are elaborated. To mitigate the patient's cancer burden, it is necessary to find an optimal modulation and frequency of treatment doses. The Stackelberg leader-follower game, adopted from the economic studies of duopoly, provides a promising framework to model the interplay between a rationally playing oncologist as a leader and the evolutionary evolving tumor as a follower. The two game types applied simultaneously to cancer therapy strategisizing can nourish each other and improve the planning of adaptive regimen. Hence, the characteristics of the Stackelberg game are mathematically studied and a preliminary dose-optimization function is presented. The applicability of the combination of the two games in the planning of cancer therapy strategies is tested with a theoretical case. The results are critically discussed from three perspectives: the biological veracity of the eco-evolutionary model, the applicability of the Stackelberg game, and the clinical relevance of the combination. The current limitations of the model are considered to invite further research on the subject.

Coral reefs form important marine ecosystems and simultaneously are at risk of deterioration due to rapidly changing environments as a consequence of human actions. Understanding their dynamics is thus important in order to be able to protect them from being destroyed. In this thesis we construct a lattice model for two life-history strategies of corals, brooders and spawners. These two strategies differ mainly in their modes of sexual reproduction, but also differences in growth and death rates as well as competitive ability are considered. We use pair approximation to help analyse the model while keeping its spatial structure. Numerical analysis is used to find the equilibria of the system as well as their stabilities, first for a single strategy and then for the two-strategy system. We find that the two strategies are able to coexist if the spawners have a higher growth rate and higher death rate and are competitively superior to brooders. This requires some reproduction over distance and a trade-off between growth and death rates. Thus we find that brooders are focusing a bigger part of their energy on long-distance reproduction, while spawners are dominating over short distances and having a higher turnover. We also find that both mutual invasibility and coexistence in the broader sense are only possible for low rates of sexual reproduction for both strategies. For higher rates of sexual reproduction we find that whichever strategy invades the lattice first will stay and the other cannot invade. Lastly we look at the effect of a change in environmental conditions, namely the acidification and temperature increase of oceans, on the two strategies and find that it affects the two strategies differently. The spawners are quickly driven to extinction by the change in environmental conditions, while brooders initially benefit from the changing conditions and only start to suffer themselves after the spawners have gone extinct.

Personalized medicine tailors therapies for the patient based on predicted risk factors. Some tools used for making predictions on the safety and efficacy of drugs are genetics and metabolomics. This thesis focuses on identifying biomarkers for the activity level of the drug transporter organic anion transporting polypep-tide 1B1 (OATP1B1) from data acquired from untargeted metabolite profiling. OATP1B1 transports various drugs, such as statins, from portal blood into the hepatocytes. OATP1B1 is a genetically polymorphic influx transporter, which is expressed in human hepatocytes. Statins are low-density lipoprotein cholesterol-lowering drugs, and decreased or poor OATP1B1 function has been shown to be associated with statin-induced myopathy. Based on genetic variability, individuals can be classified to those with normal, decreased or poor OATP1B1 function. These activity classes were employed to identify metabolomic biomarkers for OATP1B1. To find the most efficient way to predict the activity level and find the biomarkers that associate with the activity level, 5 different machine learning models were tested with a dataset that consisted of 356 fasting blood samples with 9152 metabolite features. The models included both a Random Forest regressor and a classifier, Gradient Boosted Decision Tree regressor and classifier, and a Deep Neural Network regressor. Hindrances specific for this type of data was the collinearity between the features and the large amount of features compared to the number of samples, which lead to issues in determining the important features of the neural network model. To adjust to this, the data was clustered according to their Spearman’s rank-order correlation ranks. Feature importances were calculated using two methods. In the case of neural network, the feature importances were calculated with permutation feature importance using mean squared error, and random forest and gradient boosted decision trees used gini impurity. The performance of each model was measured, and all classifiers had a poor ability to predict decreasead and poor function classes. All regressors performed very similarly to each other. Gradient boosted decision tree regressor performed the best by a slight margin, but random forest regressor and neural network regressor performed nearly as well. The best features from all three models were cross-referenced with the features found from y-aware PCA analysis. The y-aware PCA analysis indicated that 14 best features cover 95% of the explained variance, so 14 features were picked from each model and cross-referenced with each other. Cross-referencing highest scoring features reported by the best models found multiple features that showed up as important in many models.Taken together, machine learning methods provide powerful tools to identify potential biomarkers from untargeted metabolomics data.

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.

Polygenic risk scores (PRSs) estimate the genetic risk of an individual for a certain polygenic disease trait by summing up the effects of multiple variants across the genome affecting the disease risk. Currently, polygenic risk scores (PRSs) are calculated from imputed array genotyping data which is inexpensive to produce use and has standard procedures and pipelines available. However, genotyping arrays are prone to ascertainment bias, which can also lead to biased PRS results in some populations. If PRSs are utilized in healthcare for screening rare diseases, usage of whole-genome sequencing (WGS) instead of array genotyping is desirable, because also individual samples can be analyzed easily. While high-coverage WGS is still significantly more expensive than array genotyping, low-coverage whole genome sequencing (lcWGS) with imputation has been proposed as an alternative for genotyping arrays. In this project, the utility of imputed low-coverage whole-genome sequencing (lcWGS) data in PRS estimation compared to genotyping array data and the impact of the choice of imputation tool for lcWGS data was studied. Down-sampled WGS data with six different low coverages (0.1x-2x) was used to represent lcWGS data. Two different pipelines were used in genotype imputation and haplotype phasing: in the first one, pre-phasing and imputation were performed directly for the genotype likelihoods (GLs) calculated from the down-sampled data, whereas in the second one, the GLs were converted to genotype calls before imputation and phasing. In both pipelines, PRS for 27 disease phenotypes were calculated from the imputed and phased lcWGS data. Imputation and PRS calculation accuracy of the two pipelines were calculated in relation to both genotyping array and high-coverage whole-genome sequencing (hcWGS) data. In both pipelines, imputation and PRS calculation accuracy increased when the down-sampled coverage increased. The second imputation and phasing pipeline lead to better results in both imputation and PRS calculation accuracy. Some differences in PRS accuracy between different phenotypes were also detected. The results show similar patterns to what is seen in other similar publications. However, not quite as high imputation and PRS accuracy as seen in earlier studies could be attained, but possible limitations leading to lower accuracy could be identified. The results also emphasize the importance of choosing suitable imputation and phasing methods for lcWGS data and suggest that methods and pipelines designed particularly for lcWGS should be developed and published.

Genomic structural variants are large events that change the structure of the genome. These can cause changes in the functions of cells by breaking genes and genomic regulatory regions. Multiple factors are known to affect the formation of structural variants and previous studies have shown that often the sequence content in a genomic region plays a role in their formation. This study aims to characterize the sequence content around structural variant breakpoints from structural variants which have been detected from human tissue samples which have been whole genome sequenced with nanopore sequencing. The characterization was done by looking at the genomic repetitive elements found around the breakpoints, by analyzing the GC-content around the breakpoints, and by studying what kind of enriched DNA motifs were found in the sequences around the breakpoints and how these were located in these sequences. Multiple different repetitive elements were seen to occur near the breakpoint regions, and it was also observed that there were differences in what kind of repetitive elements were seen around different types of structural variants. Around the sequences of different kinds of structural variants there was also distinct differences in what kind of GC-content profiles the sequences had. In addition, various different enriched motifs were also found from the sequences and many of these showed distinct variation on how they were located around the breakpoints. These results support the previous findings showing that also here the sequence content does play a role in the formation of structural variants, but still all of the results here could not be directly explained by previous studies. In these results, it was seen that the GC-content was higher in sequences that have been affected by an event that causes structural variant formation. Also, many of the found DNA motifs were distinctly skewed around the breakpoint sequences, possibly hinting that the sequences containing these motifs would be prone to the formation of structural variants.

Sequence alignment is widely studied problem in the field of bioinformatics. The exact solution takes quadratic time to compute, and thus is not practical for long sequences. A number of heuristic approaches have been developed to conquer the quadratic time-complexity. This thesis reviews the average-case time analysis of two such heuristics, banded alignment by Ganesh and Sy in ''Near-Linear Time Edit Distance for Indel Channels'' WABI 2020, and seed-chain-extend by Shaw and Yu in ''Proving sequence aligners can guarantee accuracy in almost O(m log n) time through an average-case analysis of the seed-chain-extend heuristic'' Genome Research 2023. These heuristics reduce the quadratic average-case time complexity of the sequence alignment to log-linear. The approach of the thesis reviews is to outline the proofs of the original analysis, and provide supporting materials to aid the reader in studying the analysis. The experiments of this thesis compare four different approaches to compute the exact match anchors of the seed-chain-extend sequence alignment heuristic. A Bi-Directional Burrows-Wheeler Transformation (BDBWT), suffix tree based Mummer and Minimap2 based exact match anchors are computed. The anchors are then given to a chaining algorithm, to compare the performance of each anchoring technique. The qualities of the chains are compared using a Jaccard index applied to the sequences. The highest Jaccard index is obtained for the maximal exact match and the unique maximal exact match anchors of Mummer and BDBWT approaches. An increasing minimum length of the exact matches seem to increase the Jaccard index and reduce the running time of the chaining algorithm.

Cancer is a leading cause of death worldwide. Unlike its name would suggest, cancer is not a single disease. It is a group of diseases that arises from the expansion of a somatic cell clone. This expansion is thought to be a result of mutations that confer a selective advantage to the cell clone. These mutations that are advantageous to cells that result in their proliferation and escape of normal cell constraints are called driver mutations. The genes that contain driver mutations are known as driver genes. Studying these mutations and genes is important for understanding how cancer forms and evolves. Various methods have been developed that can discover these mutations and genes. This thesis focuses on a method called Deep Mutation Modelling, a deep learning based approach to predicting the probability of mutations. Deep Mutation Modelling’s output probabilities offer the possibility of creating sample and cancer type specific probability scores for mutations that reflect the pathogenicity of the mutations. Most methods in the past have made scores that are the same for all cancer types. Deep Mutation Modelling offers the opportunity to make a more personalised score. The main objectives of this thesis were to examine the Deep Mutation Modelling output as it was unknown what kind of features it has, see how the output compares against other scoring methods and how the probabilities work in mutation hotspots. Lastly, could the probabilities be used in a common driver gene discovery method. Overall, the goal was to see if Deep Mutation Modelling works and if it is competitive with other known methods. The findings indicate that Deep Mutation Modelling works in predicting driver mutations, but that it does not have sufficient power to do this reliably and requires further improvements.

Cancer is a dynamic and complex microevolutionary process. All attempts of curing cancer thus rely on successfully controlling also the evolving future cancer cell population. Since the emergence of drug resistance severely limits the success of many anti-cancer therapies, especially in the case of the promising targeted therapies, we need urgently better ways of controlling cancer evolution with our treatments to avoid resistance. This thesis characterizes acquired drug resistance as an evolutionary rescue and uses optimal control theory to critically investigate the rationale of aggressive maximum tolerated dose (MTD) therapies that represent the standard of care for first line treatment. Unlike the previous models of drug resistance, which mainly concentrate on minimizing the tumor volume, herein the optimal control problem is reformulated to explicitly minimize the probability of evolutionary rescue, or equivalently, maximizing the extinction probability of the cancer cells. Furthermore, I investigate the effects of drug-induced resistance, where the rate of gaining new resistant cells increases with the dose due to increased genome-wide mutation rate and non-genetic adaptations (such as epigenetic regulation and phenotypic plasticity). This approach not only reflects the biological realism, but also allows to model the cost of control in a quantifiable manner instead of using some ambiguous and incomparable penalty parameter for the cost of treatment. The major finding presented in this thesis is that MTD-style therapies may actually increase the likelihood of an evolutionary rescue even when only modest drug-induced effects are present. This suggests that significant improvements to treatment outcomes may be accomplished at least in some cases by treatment optimization. The resistance promoting properties of different anti-cancer therapies should therefore be properly investigated in experimental and clinical settings.

Traditional parametric statistical inference methods, such as maximum likelihood and Bayesian inference, cannot be used to learn parameter estimates if the likelihood is intractable, for example due to the complexity of the studied phenomenon. This can be overcome by using likelihood-free inference that is used with simulator-based models to learn parameter estimates. Also, traditional methods used in the estimation of uncertainties related to the parameter estimates typically require a likelihood function, and that is why these methods cannot be applied in likelihood-free inference. In this thesis, we present a novel way to compute confidence sets for parameter estimates obtained from likelihood-free inference using Jensen—Shannon divergence. We consider two test statistics that are based on mean Jensen—Shannon divergence and propose hypothesised asymptotic distributions for them. We test whether these hypothesised distributions can be used in the computation of confidence sets for parameter estimates obtained from likelihood-free inference, and we evaluate the produced confidence sets by studying their frequentist behaviour that is summarised with coverage probabilities. We compare this frequentist behaviour between Jensen —Shannon divergence estimates and confidence sets obtained from grid evaluation of Monte Carlo estimates and from Bayesian optimisation for likelihood-free inference (BOLFI) to the ones obtained from maximum likelihood inference with Wald’s and log likelihood-ratio confidence sets using three different models. We also use a simulator- based model with intractable likelihood to study the proposed confidence sets with BOLFI. In order to study the influence of observations on the parameter estimates and their confidence sets, we conducted these experiments with varying the number of observations. We show that Jensen—Shannon divergence based confidence sets meet the expected frequentist behaviour.