Filtyp
In this thesis, we consider a 2-Higgs-doublet model with a gauged Froggatt-Nielsen mechanism as an extension of the standard model. We assume that (i) the theory at the electroweak scale is the low energy limit of a theory defined at a scale $\Lambda_{FN}$ (ii) the Yukawa couplings at $\Lambda_{FN}$ are generated by the Froggatt-Nielsen mechanism. The first assumption is enforced by requiring that
Plasmids are circular DNA molecules found in bacteria. These molecules often contain genes coding for resilience against harmful substances such as antibiotics, in its environment. To be able to classify plasmids fast, and with small samples would be of great gain for medical care, to improve the efficiency of treatments, and to reduce the spread of antibiotics resistance. Our experimental collab
Scientists are now able to directly convert one somatic cell type into another using a procedure known as direct lineage reprogramming or transdifferentiation. In this procedure, transcription factors which are important for initiating a rewriting of the gene expressions are introduced in the cell. One specific type of reprogramming involves generating dopamine producing neurons from human adult f
Amino acids that are in close contact in a protein structure tend to co-evolve, which gives rise to sequence correlations. Direct coupling analysis (DCA) is a method for predicting such contacts directly from sequence correlations, without assuming any prior knowledge of structures. To this end, sequence correlations are modeled using an Ising-like ansatz, whose couplings are determined through an
Artificial neural networks have been used to solve different problems, one being survival analysis of medical data. For survival analysis, the main interest is often how samples become sorted by their outputs, which makes survival analysis a rank based problem. A rank based error function lacks a gradient, which makes gradient descent based training difficult. A genetic training algorithm, based o
Convolutional Neural Networks (CNNs) and pre-trained word embeddings have revolutionized the field of Natural Language Processing (NLP) during the last years. In this project, CNNs are used on top of the Word2Vec word representation for a sentence classification task on medical research articles. Both individual networks for each category as well as a combined classification network are optimized
In the field of computational biology, fluorescence microscopy images often constitute the input source of information. The process of binarization of raw images to delineate interesting objects requires image segmentation into signal and background pixels. Several methods to perform image segmentation exist, the Otsu method being a popular unsupervised example. The Otsu method's lack of pro
We define a mathematical method that, given the form of an interaction network, defines an energy as a function of the state of the network. When combined with a suitable entropy, it yields a free energy that gives rise to the dynamics of the network, allowing solutions and optimal paths to be efficiently found. This is done primarily with the gene regulatory networks governing cell reprogramming
Direct coupling analysis (DCA) models correlations in sets of related (homologous) protein sequences using a Potts-like spin model ansatz. From the couplings of the Potts model, derived by inverse statistical mechanics, residue-pair contacts in the 3D structure of the protein are predicted. In this thesis, this approach is applied to structures from the HP model on a square lattice. All HP sequenc
The recently introduced capsule networks have already shown much success on image classification tasks. In this thesis, capsule networks are described and applied to a binary classification task, namely that of classifying hotspots from bone scintigraphy scans. The performance of capsule networks on this task is compared to that of convolutional neural networks. The results indicate that there is
Genetic algorithms are complex constructs often used as heuristic search methods in contexts ranging from combinatorial optimisation to in silico evolution. They draw inspiration from the principles of biological evolution by utilizing the concepts of mutation, reproduction and selection in order to improve a population of solutions. The solutions are often represented as abstract data sequences,
In this project, we compare two error functions for the purpose of training artificial neural networks on heavily censored data (data where key information is missing). J. Kalderstam et al. has shown that it is possible to train artificial neural networks directly on Harrell's C index \cite{Harrell} using genetic algorithms. He has also investigated the possibilities of improving the performa
The growth of tissues and organs in plants is governed by the morphogen auxin coupled with the membrane protein PIN, which together generate patterns that guide development. Systems of this kind have been studied extensively in experiments and computational system biology models. This thesis builds on that work by introducing a stochastic version of these models to examine differences between stoc
During the last few years, crowding effects on the physics of proteins has become an increasingly popular topic of research. This is is because most biological processes involving proteins naturally take place in a crowded environment, e.g. in the cellular environment where macromolecules may occupy 30% of the volume. One such biological process would be the formation of amyloid aggregates, which
The generation of induced pluripotent stem (iPS) cells from differentiated cells is a process of great scientific interest due to the medical potential of such cells. This process (called 'reprogramming' of cells) is however notoriously inefficient and still not very well understood. Therefore studies that can help us understand the interactions and mechanisms governing the reprogramming p
Artificial Neural Networks (ANNs) are widely used information processing algorithms based roughly on biological neural networks. These networks can be trained to find complex patterns in datasets and to produce certain output signals given a set of input signals. A key element of ANNs are their so-called activation functions, which control the signal strengths between the artificial neurons in a n
Outlier detection in high-dimensional data is a complex task, useful in many fields. One major application is in healthcare, where the high-dimensional healthcare registers can be used to detect patterns related to the medical behaviour and state of the population. In this project, neural network based autoencoders were used to study the Covid-19 vaccination campaign, where it was trained on healt
This thesis examines the state-of-the-art 2D super-resolution technique alternating descent conditional gradient (ADCG) method's ability to accurately localize fluorophores in diffraction-limited single molecule images (SMI) and analyze the impact of pre-processing and post-processing modules on ADCG's fluorophore localization. A synthetic dataset obtained from the 2013 Grand Challenge loc
