The ability to predict is one of the hallmarks of successful theories. Historically, the predictive power of biology has lagged behind disciplines like physics because the biological world is complex, challenging to quantify, and full of exceptions. However, in recent years the amount of available data has expanded exponentially and biological predictions based on this data become a possibility. The functional gene network is a quantitative way to integrate this data and a useful framework for making biological predictions. This study demonstrates that functional networks capture real biological insight and uses the network to predict both subcellular protein localization and the phenotypic outcome of gene knockouts. Furthermore, I use the functional network to evaluate genetic modules shared between diverse organisms that lead to orthologous phenotypes, many that are non-obvious. I show that the successful predictions of the functional network have broad applicability and implications that range from the design of large-scale biological experiments to the discovery of genes with potential roles in human disease. / text

This study examined factors related to word list performance predictions made by younger and older adults. A performance prediction is an estimate made prior to being exposed to the material that is studied for a specific task. The current study examined the age differences in a sample of 59 older adults (M = 76.83 years old, SD = 8.28) and 51 younger adults (M = 21.19 years old, SD = 3.22) on performance predictions for both an immediate and delayed word recall task. Memory self-efficacy and other self-rating measures were not found to influence immediate or delayed predictions. A repeated measures ANOVA revealed that older adults improved in absolute accuracy from immediate to delayed prediction whereas younger adults became less accurate. The results suggest that all metamemory skills do not deteriorate with age, as the older adults were capable of monitoring their memory accurately based on previous performance.

Metabolic networks are represented by the set of metabolic pathways. Metabolic pathways are a series of chemical reactions, in which the product from one reaction serves as the input to another reaction. Many pathways remain incompletely characterized, and in some of them not all enzyme components have been identified. One of the major challenges of computational biology is to obtain better models of metabolic pathways. Existing models are dependent on the annotation of the genes. This propagates error accumulation when the pathways are predicted by incorrectly annotated genes. Pairwise kernel frameworks have been used in supervised learning approaches, e.g., Pairwise Support Vector Machines (SVMs), to predict relationships among two pairs of entities. Rational kernels are based on transducers to manipulate sequence data, computing similarity measures between sequences or automata. Rational kernels take advantage of the smaller and faster representation and algorithms of weighted finite-state transducers. They have been effectively used in problems that handle large amount of sequence information such as protein essentiality, natural language processing and machine translations. We propose a new framework, Pairwise Rational Kernels (PRKs), to manipulate pairs of sequence data, as pairwise combinations of rational kernels. We develop experiments using SVM with PRKs applied to metabolic pathway predictions in order to validate our methods. As a result, we obtain faster execution times with PRKs than other kernels, while maintaining accurate predictions. Because raw sequence data can be used, the predictor model avoids the errors introduced by incorrect gene annotations. We also obtain a new type of Pairwise Rational Kernels based on automaton and transducer operations. In this case, we define new operations over two pairs of automata to obtain new rational kernels. We also develop experiments to validate these new PRKs to predict metabolic networks. As a result, we obtain the best execution times when we compare them with other kernels and the previous PRKs.

People share a big part of their lives and opinions on platforms such as Facebook and Twitter. The companies behind these sites do their absolute best to collect as much data as possible. This data could be used to extract opinions in many different ways. Every company, organization or public person is probably curious on what is being said about them right now. There are also areas where opinions are related to the outcome of an event. Examples of such events are presidential elections or the Eurovision Song Contest. In these events, peoples’ votes will directly reflect the outcome of the elections or contests. We have developed a simplistic prototype that is able to predict the result of the Eurovision Song Contest using sentiment analysis on tweets. The prototype collects tweets about the event, performs sentiment analysis, and uses different filters to predict the ranks of the contestants. We evaluted our results with the actual voting results of the event and found a Pearson correlation of approximately 0.65. With more time and resources we believe that it is possible to create a highly accurate prediction model. It could be used in lots of different contexts. Politicians and their parties could use it to evaluate their campaigns. The press could use it to create more interesting news reports. Companies would be able to investigate their brand appreciation. A system like this could be used in many different fields.

I present behavioural data contributing to existing research that (implicit) self-predictions are overly reliant on current intentions at the time of the decision (Koehler & Poon, 2006). Results are consistent with previous findings that self-predictions are often insensitive to translatability cues and overly influenced by desirability cues. We show that although participants typically benefit from a reminder, it is undervalued at the time of the decision (Experiment 1 & 3a) as participants are not willing to pay for a reminder service, unless it is offered free of charge (Experiment 2). Our findings also show that participants fail to incorporate temporal delay sufficiently in their opt-in decisions, even though temporal delay was found to be a significant predictor return behaviour (Experiments 1, 2 & 3b). Instead, decisions were found to be highly influenced by desirability factors (Experiments 1 & 2) which were not significant predictors of task completion. Finally, using a construal manipulation intended to induce participants to think about the decision options in either a concrete or abstract way influenced decisions (Experiment 3a), and subsequently influenced how much participants benefitted from the reminder in task completion (Experiment 3b).

The roads in our society are in a state of constant degradation. The reasons for this are many, and therefore constructed to have a certain lifetime before being reconstructed. To minimize the cost of maintaining the important transport road network high quality prediction models are needed. This report presents new models for flexible pavement structures for initiation and propagation of fatigue cracks in the bound layers and rutting for the whole structure. The models are based on observations from the Swedish Long Term Pavement Performance (LTPP) database. The intention is to use them for planning maintenance as part of a pavement management system (PMS). A statistical approach is used for the modeling, where both cracking and rutting are related to traffic data, climate conditions, and the subgrade characteristics as well as the pavement structure. / <p>QC 20130325</p>

Understanding how genetic variation propagate to differences in phenotypes in individuals is an ongoing challenge in genetics. Genome-wide association studies have allowed for the identification of many trait-associated genomic loci. However, they are limited in their inability to explain the altered cellular mechanism. Genetic variation can drive disease by altering a range of mechanisms, including signalling networks, TF binding, and protein folding. Understanding the impact of variants on such processes has key implications in therapeutics, drug development, and more. This thesis aims to utilise computational predictors to shed light on how cellular mechanisms are altered in the context of genetic variation and better understand how they drive both molecular and organism-level phenotypes. Many binding events in the cell are mediated by short stretches of sequence motifs. The ability to discover these underlying rules of binding could greatly aid our understanding of variant impact. Kinase–substrate phosphorylation is one of the most prominent post-translational modifications (PTMs) which is mediated by such motifs. We first describe a computational method which utilises interaction and phosphorylation data to predict sequence preferences of kinases. Our method was applied to 57% of human kinases capturing known well-characterised and novel kinase specificities. We experimentally validate four understudied kinases to show that predicted models closely resemble true specificities. We further demonstrate that this method can be applied to different organisms and can be used for other phospho-recognition domains. The described approach allows for an extended repertoire of sequence specificities to be generated, particularly in organisms for which little data is available. TF-DNA binding is another mechanism driven by sequence motifs, which is key for the tight regulation of gene expression and can be greatly altered by genetic variation. We have comprehensively benchmarked current methods used to predict non-coding variant effects on TF-DNA binding by employing over 20,000 compiled allele-specific ChIP-seq variants across 94 TFs. We show that machine learning-based approaches significantly outperform more rudimentary methods such as the position weight matrix. We further note that models for many TFs with distinct binding specificities were unable to accurately assess the impact of variants. For these TFs, we explore alternative mechanisms underlying TF-binding, such as methylation, co-operative binding, and DNA shape that drive poor performance. Our results demonstrate the complexity of predicting non-coding variant effects and the importance of incorporating alternative mechanisms into models. Finally, we describe a comprehensive effort to compile and benchmark state-of-the-art sequence and structure-based predictors of mutational consequences and predict the effect of coding and non-coding variants in the reference genomes of human, yeast, and E. coli. Predicted mechanisms include the impact on protein stability, interaction interfaces, and PTMs. These variant effects are provided through mutfunc, a fast and intuitive web tool by which users can interactively explore pre-computed mechanistic variant impact predictions. We validate computed predictions by analysing known pathogenic disease variants and provide mechanistic hypotheses for causal variants of unknown function. We further use our predictions to devise gene-level functionality scores in human and yeast individuals, which we then used to perform gene-phenotype associations and uncover novel gene-phenotype associations.

Gene-finding programs available over the Internet today are shown to be nothing more than guides to possible coding regions in the DNA. The programs often do incorrect predictions. The idea of combining a number of different gene-finding programs arised a couple of years ago. Murakami and Takagi (1998) published one of the first attempts to combine results from gene-finding programs built on different techniques (e.g. artificial neural networks and hidden Markov models). The simple combinations methods used by Murakami and Takagi (1998) indicated that the prediction accuracy could be improved by a combination of programs. In this project artificial neural networks are used to combine the results of the three well-known gene-finding programs GRAILII, FEXH, and GENSCAN. The results show a considerable increase in prediction accuracy compared to the best performing single program GENSCAN

A database, AmyProt, was developed that collated details of 32 human amyloid proteins associated with disease and 488 associated mutations and polymorphisms, of which 316 are classified as amyloid. A detailed profile of the mutations was developed in terms of location within domains and secondary structures of the proteins and functional effects of the mutations. The data was used to test the hypothesis that mutations enhance amyloidosis in human amyloid proteins have distinctive characteristics, in terms of specific location within proteins and physico-chemical characteristics, which differentiate them from non-amyloid forming polymorphisms in amyloid proteins and from disease mutations and polymorphisms in non-amyloid disease linked proteins. The aim was to use these characteristics to train a prediction algorithm for amyloid mutations that will provide a more accurate prediction than current general disease prediction tools and amyloid prediction tools that focus on aggregating regions. 66 location specific features and changes upon mutation of 366 amino acids propensities, derived from the amino acid index database AAindex, were analysed. A significant proportion of mutations were located with aggregating regions, however the majority of mutations were not associated with these regions. An analysis of motifs showed that amyloid mutations had a significant association with transmembrane helix motifs such as GxxxG. Statistical analysis of substitutions mutations, using substitution matrices, showed that amyloid mutations have a decrease in α-helix propensity and overall secondary structure propensity compared to the disease mutations and disease and amyloid polymorphisms. Machine learning was used to reduce the large set of features to a set of 18 features. These included location near transmembrane helices, secondary structure features; transmembrane and extracellular domains and 4 amino acid propensities: knowledge-based membrane propensity scale from 3D helix; α-helix propensity; partition coefficient; normalized frequency of coil. The AmyProt mutations and non-amyloid polymorphisms were used to train and test the novel amyloid mutation prediction tool, AmyPred, the first tool developed purely to predict amyloid mutations. AmyPred predicts the amyloidogenicity of mutations as a consensus by majority vote (CMV) and mean probability (CMP) of 5 classifiers. Validation of AmyPred with 27 amyloid mutations and 20 non-amyloid mutations from APP, Tau and TTR proteins, gave classification accuracies of 0.7/0.71 (CMV/CMP) and with an MCC of 0.4 (CMV) and 0.41 (CMP). AmyPred out performed other tools such as SIFT (0.37) and PolyPhen (0.36) and the amyloid consensus prediction tool, MetAmyl (0.13). Finally, AmyPred was used to analyse p53 mutations to characterize amyloid and non-amyloid mutations within this protein.

The diploma thesis deals with the situation on senior tourism market. The aim of the diploma thesis is to characterize seniors on tourism market, describe the current situation in senior tourism market and define the most important characteristics of senior customer,including possible segmentation. The diploma thesis is divided into six chapters.