Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Density functional theory (DFT) can improve machine-learning predictions of nucleophilic aromatic substitution (SNAr) by supplying mechanistic information about the reaction’s transition state. In a 2021 study, Jorner and co-authors combined DFT-derived descriptors with Gaussian Process Regression (GPR) to predict experimental activation barriers. They reported a mean absolute error of 0.77 kcal mol−1 on an external test set and 86% top-1 accuracy for selectivity predictions on patent data. Those results apply to the study’s SNAr tasks, not to chemical reactions in general.
What DFT contributes to the model
DFT is a quantum-chemical method used to calculate properties of molecular systems. In this work, it was used to model reaction mechanisms and transition states. The resulting reaction information was supplied to GPR, a machine-learning approach, as descriptors to help predict experimental activation barriers.
The distinction matters: the model did not simply learn from reactant and product structures, nor did it use DFT to predict a computed barrier and treat that as the final answer. Its target was an experimental activation barrier, while DFT-based mechanistic information enriched the model’s input. The authors describe this as a hybrid of mechanistic modelling and machine learning. Read the primary paper in Chemical Science.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →How accurately did it predict SNAr barriers?
Jorner, Brinck, Norrby and Buttar report a mean absolute error (MAE) of 0.77 kcal mol−1 on an external test set. MAE is the average absolute difference between predicted and observed values, so this figure summarizes barrier-prediction error across that evaluation set; it does not mean every prediction was within 0.77 kcal mol−1.
#1 Best Overall
The external evaluation is useful evidence that the approach could predict beyond the data used to fit it. But the reported performance remains specific to the paper’s SNAr dataset, modelling choices and validation design. It should not be read as an accuracy guarantee for other reaction families, experimental conditions or datasets.
What the selectivity result means
The authors also used predicted barriers to assess reaction selectivity. On patent reaction data, they report 86% top-1 accuracy for regioselectivity and chemoselectivity predictions. Top-1 accuracy asks whether the model’s highest-ranked outcome matches the recorded outcome. The paper says the model was not trained explicitly for this selectivity task, making the result a test of what its barrier predictions could support beyond direct barrier estimation.
Rank #2
- Use scikit-learn to track an example ML project end to end
- Explore several models, including support vector machines, decision trees, random forests, and ensemble methods
- Exploit unsupervised learning techniques such as dimensionality reduction, clustering, and anomaly detection
- Dive into neural net architectures, including convolutional nets, recurrent nets, generative adversarial networks, autoencoders, diffusion models, and transformers
- Use TensorFlow and Keras to build and train neural nets for computer vision, natural language processing, generative models, and deep reinforcement learning
This is a separate evaluation from the external barrier test: it uses patent reaction data and a categorical ranking metric rather than measuring average numerical error in activation barriers. The 86% figure therefore does not describe barrier accuracy or establish the same performance for other selectivity datasets.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsWhy the authors emphasize modest data needs
Experimental kinetic measurements can be difficult to assemble in large, consistent datasets. The authors present their method as appropriate for low-data reaction classes and say that 100–150 rate constants may be enough. That is the paper’s stated regime, not a universal minimum or a guarantee that any reaction class will be learnable from that many measurements.
Rank #3
The rationale is that mechanistic descriptors can provide chemically informative inputs where a model trained only on a large collection of examples might be constrained by data availability. The approach still depends on the quality and relevance of those descriptors: machine learning cannot reliably compensate for information that fails to represent the chemistry being modelled.
Uncertainty and appropriate use
The paper reports prediction error bars, which can help users assess risk rather than relying on a point prediction alone. Such estimates are useful when deciding whether a prediction is sufficiently reliable to inform a follow-up experiment. They do not remove uncertainty or substitute for experimental validation, particularly when applying the model outside the chemical space and validation tasks studied.
Rank #4
For researchers evaluating this kind of method, the key questions are whether the target is an experimental barrier or another quantity, whether inputs encode mechanistic or transition-state information, how much kinetic data are available, and whether validation is external to model fitting. For selectivity claims, the metric and dataset also matter. The 2021 study offers an example of combining these elements for SNAr; it does not establish a general-purpose predictor for arbitrary transformations.
Study and publication details
The primary paper, “Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies,” is by Kjell Jorner, Tore Brinck, Per-Ola Norrby and David Buttar. It appeared in Chemical Science, volume 12, pages 1163–1175 (2021), and was first published online on 5 November 2020. The Royal Society of Chemistry article page provides the paper and its reported methods and results.
Best Value
Chemistry World’s 6 January 2021 report offers commentary from external researchers. Those comments provide context on the promise of transition-state descriptors and the importance of good inputs; they are expert opinions, not additional experimental validation of the model.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
