MPF-BML

MPF-BML infers maximum-entropy Boltzmann machine models from sequence and other high-dimensional biological data using the minimum probability flow algorithm to learn correlation patterns and estimate parameters such as viral protein fitness landscapes.

Key Features:

• Minimum Probability Flow Estimation: Uses the minimum probability flow algorithm for fast and consistent parameter estimation.
• Boltzmann Machine / Maximum Entropy Inference: Performs parameter inference within a Boltzmann Machine Learning framework to recover maximum entropy models.
• Scalability for High-Dimensional Data: Targets large-scale, high-dimensional applications where traditional inference methods face computational challenges.
• Correlation Pattern Learning: Recovers correlation structures within empirical datasets to inform model parameters.
• Application to Viral Protein Fitness Landscapes: Has been applied to genetic sequence data to estimate fitness landscapes for surface proteins of HIV and HCV.

Scientific Applications:

• Maximum-Entropy Model Inference: Infers maximum-entropy models to characterize statistical dependencies in biological data.
• Genetic Sequence Analysis: Analyzes genetic sequence datasets to extract parameter estimates relevant to evolutionary and functional studies.
• Fitness Landscape Estimation: Estimates fitness landscapes for viral surface proteins, exemplified by applications to HIV and HCV sequence data.

Methodology:

Parameter inference is performed by applying the minimum probability flow approach within a Boltzmann Machine Learning framework to fit maximum entropy models for large-scale data.