t-Distributed Stochastic Neighbor Embedding Method with the Least Information Loss for Macromolecular Simulations

Area of Science:

• Computational Biology
• Biophysics
• Structural Bioinformatics

Background:

• Dimensionality reduction is crucial for analyzing complex macromolecule simulations.
• Existing methods like PCA and t-ICA often lose vital structural information.
• Distinguishing functionally important states requires methods that preserve conformational details.

Purpose of the Study:

• To introduce t-distributed stochastic neighbor embedding (t-SNE) as a superior dimensionality reduction technique for macromolecule simulations.
• To evaluate t-SNE's ability to minimize structural information loss.
• To demonstrate t-SNE's effectiveness in identifying distinct functional states in protein dynamics.

Main Methods:

• Application of t-distributed stochastic neighbor embedding (t-SNE) to molecular dynamics simulations.
• Comparison of 1D and 2D t-SNE models against traditional methods (PCA, t-ICA, CVs).
• Analysis of a model allosteric protein system to assess state distinguishability and mechanistic insights.

Main Results:

• t-SNE demonstrated minimal structural information loss compared to conventional methods.
• 1D and 2D t-SNE models effectively distinguished important functional states of the allosteric protein.
• t-SNE projections provided clear visual and quantitative data on protein transition mechanisms.

Conclusions:

• t-SNE is a powerful tool for dimensionality reduction in biomacromolecule simulations.
• It offers enhanced capabilities for analyzing protein conformational dynamics and functional states.
• t-SNE provides valuable insights into transition mechanisms not easily obtained with other techniques.