Rapid Generation of Transition-State Conformer Ensembles via Constrained Distance Geometry.

Area of Science:

• Computational chemistry
• Catalysis
• Machine learning

Background:

• Accurate modeling of chemical reactions requires transition-state (TS) conformer ensembles.
• Established methods like CREST and GOAT are computationally expensive, hindering large-scale applications.
• Efficient TS conformer generation is crucial for computational catalyst design and machine learning dataset creation.

Purpose of the Study:

• Introduce racer^TS, a novel method for rapid transition-state conformer ensemble generation.
• Benchmark racer^TS against existing methods (CREST, GOAT) in terms of efficiency and accuracy.
• Assess the utility of racer^TS for generating datasets for computational chemistry and machine learning.

Main Methods:

• Developed racer^TS based on constrained distance geometry.
• Generated TS conformer ensembles for 20 diverse reactions.
• Optimized selected conformers using DFT and evaluated performance metrics: cost, exhaustiveness, validity, and low-energy accuracy.

Main Results:

• racer^TS covers the conformer space comparably to CREST and slightly less than GOAT.
• racer^TS demonstrates improved validity of DFT-optimized TSs and sufficient accuracy in low-energy regions (median error 0.17 kcal/mol).
• Significantly reduced computational wall-time compared to existing methods.

Conclusions:

• racer^TS is a highly efficient TS conformer ensemble generator for computational chemistry.
• Enables rapid TS conformer sampling and the creation of valuable datasets for machine learning.
• Facilitates the discovery of novel and sustainable catalysts.