On geometric probability distributions on the torus with applications to molecular biology (Q2323940)

Motivated by the protein folding problem in biochemistry, the author investigates properties of univariate and multivariate distributions on the torus which provide alternatives to the von Mises and wrapped normal distributions. The univariate Inverse Stereographic Normal Distribution (ISND) with parameters \(\mu\in\mathbb{R}\) and \(\sigma>0\) has density function given by \[ f(\theta)=\sqrt{\frac{1}{2\pi}}\frac{1}{(1+\cos\theta)\sigma}\exp\left\{-\frac{1}{2\sigma^2}\left(\frac{\sin\theta}{1+\cos\theta}-\mu\right)^2\right\}\,, \] for all \(\theta\in[-\pi,\pi)\) on the circle. The multivariate version of this distribution is defined similarly. Properties of these families of distributions are investigated, including closure under marginalization and conditioning, and conditions under which the distribution is unimodal or multimodal. It is also shown that these distributions can be close to von Mises and wrapped normal distributions. Aspects of statistical inference (including point estimation, interval estimation and hypothesis testing) are also discussed for these ISNDs. The inverse stereographic moment of order \(k\) is defined for a random variable on the circle with density function \(f\) by \[ \int_{-\pi}^\pi\left(\frac{\sin\theta}{1+\cos\theta}\right)^kf(\theta)\,d\theta\,. \] Using this definition, the corresponding moment generating function is defined and used to formulate a central limit-type result for such distributions on the torus. Finally, numerical experiments illustrate the above results, and applications of these distributions to molecular biology are discussed.