NeuralTSNE: A Python Package for the Dimensionality Reduction of Molecular Dynamics Data Using Neural Networks

Area of Science:

• Computational chemistry
• Machine learning
• Data science

Background:

• Unsupervised machine learning is increasingly used for molecular dynamics (MD) data analysis.
• Dimensionality reduction techniques are crucial for extracting insights from high-dimensional MD trajectories.
• Standard t-distributed stochastic neighbor embedding (t-SNE) is popular, but parametric versions show improved performance.

Purpose of the Study:

• To introduce NeuralTSNE, a Python package implementing parametric t-SNE.
• To provide an accessible tool for analyzing molecular dynamics data using advanced dimensionality reduction.
• To leverage neural networks for enhanced performance in t-SNE applications.

Main Methods:

• Implementation of parametric t-distributed stochastic neighbor embedding (t-SNE) using PyTorch and PyTorch Lightning.
• Development of a user-friendly Python package, NeuralTSNE.
• Application of the package to molecular dynamics (MD) data analysis.

Main Results:

• NeuralTSNE provides an effective implementation of parametric t-SNE.
• The package demonstrates superior performance in dimensionality reduction compared to standard t-SNE.
• NeuralTSNE facilitates the analysis of complex molecular dynamics data.

Conclusions:

• NeuralTSNE is a valuable and easy-to-use tool for researchers in molecular dynamics.
• Parametric t-SNE implemented in NeuralTSNE offers enhanced capabilities for MD data analysis.
• The package supports both module import and command-line usage for flexibility.