Bioinformatics Project - 7.5 ECTS

Numpyro implementations of univariate and bivariate Von Mises distributions

The goal of the current project was to implement the univariate and bivariate von Mises distributions in the probabilistic programming language Numpyro in Python 3. The alpha release of Numpyro uses JAX for automatic diferentiation and JIT compilation to TPU, GPU & CPU. Due to its recency, many important statistical distributions and other important features are, however, still lacking. As both univariate and bivariate von Mises distributions are widely used in modelling the distribution of torsional angles and of amino acid backbone chains, these implementations will have direct uses especially in protein structure prediction. Although the current report focuses on the use of circular statistics in bioinformatics, it should be noted that these distributions are also commonly used in e.g. physics & astronomy. The project report is available under files in this GitHub repository.