The Gaussian-linear hidden Markov model: A Python package.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Hidden Markov Models (HMMs) are widely used in neuroscience for analyzing complex time-series data.
• Existing HMMs often have limitations in flexibility and scalability for diverse neural data types.

Purpose of the Study:

• To introduce the Gaussian-Linear Hidden Markov Model (GLHMM) as a generalized framework for neural data analysis.
• To provide a flexible and scalable computational toolbox for uncovering brain-behavior associations.

Main Methods:

• Developed the GLHMM, a generalization of HMMs, using linear regression to parameterize Gaussian state distributions.
• Implemented a Python toolbox utilizing stochastic variational inference for efficient analysis of large datasets.
• Demonstrated applicability across various neuroimaging and physiological data types (fMRI, LFP, ECoG, MEG, pupillometry).

Main Results:

• The GLHMM framework accommodates unsupervised, encoding, and decoding models within a unified approach.
• The toolbox enables robust statistical testing and out-of-sample prediction for characterizing brain-behavior relationships.
• Efficient computation allows handling of large-scale datasets in reasonable time.

Conclusions:

• The GLHMM offers a powerful and versatile tool for advancing neuroscience research.
• The associated Python toolbox democratizes advanced statistical modeling for brain-behavior association studies.
• GLHMM is suitable for a broad range of experimental paradigms and data modalities.