Flashcard library · Chemistry · PaperFren
ML plus transition states predict SNAr barriers
A Gaussian-process model that includes SNAr transition-state features predicts experimental activation energies with 0.77 kcal mol−1 MAE, beating raw DFT.
Keep what you just learned
Save these 11 cards to your NoteFren library. Your first review starts right away, and each card links back to the study it came from.
What reaction class is the prototype?
Nucleophilic aromatic substitution (SNAr).
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
What test-set MAE did GPR reach?
0.77 kcal mol−1, below 1 kcal mol−1 chemical accuracy.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
How does raw DFT compare?
MAE 2.93 kcal mol−1, equivalent to guessing the training mean.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
What R2, MAE, and RMSE on the external set?
R2 = 0.93, MAE = 0.77, RMSE = 1.01 kcal mol−1.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
What selectivity accuracy was achieved without training on that task?
86% top-1 (85% regio, 87% chemo) on patent data.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
Which regressor won?
Gaussian process regression with a Matern 3/2 kernel.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
When are hybrid TS-feature models most useful?
With about 50–150 samples.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
What extra features do hybrid models add?
Transition-state features plus traditional physical-organic descriptors.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
How did linear regression and random forests score in CV?
LR MAE 1.20 and RF 0.98 kcal mol−1 versus GPR ~0.80.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
What coverage do prediction intervals have on the test set?
99%, so uncertainty estimates are well calibrated.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified
What still holds back hybrid models?
Difficulty of computing transition states reliably.
Source: Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies · DOI · quote verified