DONKEY: A Flexible and Accurate Algorithm for Clustering

Area of Science:

• Computational chemistry
• Chemical physics
• Data science

Background:

• Analyzing complex molecular dynamics simulations, especially photoexcited states, presents significant data challenges.
• Existing clustering algorithms often require parameter tuning, introducing bias and limiting applicability to varied datasets.

Purpose of the Study:

• To introduce a novel, parameter-free clustering algorithm for analyzing multidimensional temporal data from nonadiabatic trajectory-based simulations.
• To enhance the identification of distinct reaction pathways in photoexcited molecular systems.

Main Methods:

• Variable kernel density estimation to approximate probability density functions.
• Assignment of data points to local maxima representing cluster centers.
• Merging of clusters to overcome minor density fluctuations, ensuring robust separation.

Main Results:

• The algorithm demonstrates superior performance compared to conventional methods on synthetic datasets.
• Successfully applied to the photoexcited dynamics of the norbornadiene ⇌ quadricyclane molecular photoswitch.
• Identified distinct reaction pathways, showcasing its practical utility.

Conclusions:

• The proposed clustering algorithm offers an accurate and flexible tool for analyzing complex simulation data.
• Its parameter-free nature reduces bias and enhances applicability across diverse scientific domains.
• Facilitates deeper insights into reaction mechanisms in photochemistry and molecular dynamics.