Marina Axelson-Fisk

Professor, Mathematical Sciences

Marina Axelson-Fisk is a Professor in Mathematical Statistics. She works mainly within the field of Bioinformatics, with particular focus on stochastic models and algorithms for comparative genomics and cross-species gene finding. She was, among other things, involved in the initial comparative analyses of the human, mouse and rat in the Mouse and the Rat Sequencing Consortia. She is an associate member of the Linnaeus Centre for Marine Evolutionary Biology (CeMEB), where she is involved in the bioinformatics analysis of a number of newly sequenced Swedish coastal organisms.

For more information, see my homepage at
​Mathematical statistics, probability theory and bioinformatics.
​Chemistry and Molecular Biology, University of Gothenburg
Medical Biochemistry, University of Gothenburg
Biological and environmental sciences, University of Gothenburg
Applied Mechanics, Chalmers
SAFER Vehicle and traffic safety centre, Chalmers
Mathematics, University of California, Berkeley
Chalmers Area of Advance, Life Sciences

Bioinformatics project: Comparative gene prediction in bacteria

This project concerns the development of probabilistic models and computational tools for comparative genomics in bacteria. The PhD student is expected to have a solid background in mathematical statistics with particular interest in stochastic processes. Also, some experience in computer programming would be preferred. Background knowledge in biology is not required, but is of course an advantage. The focus of the project will be on Hidden Markov Models (HMMs) and Conditional Random Fields (CRFs).
Computational gene prediction involves the identification of signals and statistical sequence patterns that distinguish functional from non-functional elements in DNA sequence. Some of the most successful gene prediction tools are based on hidden Markov models (HMMs), which are highly accurate, with a straightforward probabilistic interpretation, and a convenient framework for model validation and parameter training.
HMMs consists of two interrelated processes, one Markov process that is hidden from the observer and that jumps between states in a state space, and an observed process that is a function of the underlying state. HMMs has been successfully used in many areas of bioinformatics, in particular in computational gene prediction, where the observed process is the DNA sequence, and the hidden process is the segmentation into functional and non-functional elements in the DNA.
CRFs can be seen as a generalization of HMMs that uses discriminative training as opposed to the generative mode of HMMs. CRFs are more flexible than HMMs as they can be conditioned on a set of global “features”, that can include both dependent and independent, and both probabilistic and non-probabilistic components. The flexibility comes at the price of computational complexity, however, and one focus will be to investigate different ways of handling this.
Comparative gene prediction utilizes the fact that functional elements in the genome evolve at a much lower rate than non-functional DNA. Therefore, sequence comparisons between evolutionary related species can strengthen the signal of important functional regions in the genome. The choice of bacteria as model organisms has several advantages: their smaller and compact genomes provide a suitable test-bed for the development of computationally complex tools, the bacterial sequences come in abundance and with high diversity in a wide span of evolutionary distances, and the analysis of bacteria has important health applications in epidemiology and antibiotic resistance.

Page manager Published: Fri 22 Apr 2022.