Mocapy++

Mocapy++ implements parameter learning and inference for dynamic Bayesian networks (DBNs) and provides directional statistics and specialized probability distributions for modeling biomolecular structures such as protein and RNA coordinates at the atomic level.

Key Features:

• Implementation: Implemented in C++.
• DBN architectures: Supports parameter learning and inference across multiple dynamic Bayesian network architectures.
• Node types: Supports discrete, multinomial, Gaussian, and Poisson node distributions.
• Directional statistics: Supports directional distributions including the Kent distribution on the sphere, the Von Mises distribution, and the bivariate von Mises distribution on the torus.
• Inference and learning algorithms: Uses Gibbs sampling and Stochastic-EM (Expectation-Maximization) for parameter estimation.
• Atomic-level modeling support: Provides probabilistic primitives and distributions aimed at modeling protein and RNA structures at the atomic level.

Scientific Applications:

• Probabilistic biomolecular modeling: Construction of probabilistic models of biomolecular structures using DBNs and directional distributions.
• Protein structure modeling: Modeling of protein coordinates and orientations at atomic resolution using Kent and von Mises distributions.
• RNA structure modeling: Modeling of RNA atomic-level structure and orientations with specialized directional statistics.
• Angular and directional data analysis: Analysis of angles, directions, and orientations in biomolecular datasets.

Methodology:

Inference and parameter estimation are performed using Gibbs sampling and Stochastic-EM (Expectation-Maximization) within dynamic Bayesian networks, employing directional distributions such as the Kent distribution and the bivariate von Mises distribution.