Toward Chemical Accuracy in Predicting Enthalpies of Formation with General-Purpose Data-Driven Methods

Zheng, Peikun; Yang, Wudi; Wu, Wei; Isayev, Olexandr; Dral, Pavlo O.

doi:10.1021/acs.jpclett.2c00734

Cited by 27 publications

(37 citation statements)

References 76 publications

(142 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The three examples taken from the literature, as well as the ones presented in a former study, 18 show that designing prediction uncertainties is a demanding process, and that it is important to use the right validation tools. We have seen for instance that some diagnostics used in the literature, such as cumulative MAE curves (confidence curves) 4 or R 2 regression coefficients between |E| and 47,48 should not be used to assess calibration. Both metrics detect that large errors are associated with large uncertainties, but they do not test if the scaling is correct.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Prediction uncertainty validation for computational chemists

Pernot

2022

Preprint

View full text Add to dashboard Cite

Validation of prediction uncertainty (PU) is becoming an essential tool for modern computational chemistry. Designed to quantify the reliability of predictions in meteorology, the calibration-sharpness (CS) framework is now widely used to optimize and validate uncertainty-aware machine learning (ML) methods. However, its application is not limited to ML and it can serve as a principled framework for any PU validation. The present article is intended as a step-by-step introduction to the concepts and techniques of PU validation in the CS framework, adapted to the specifics of computational chemistry. The presented methods range from elementary graphical checks to more sophisticated ones based on local calibration statistics. The concept of tightness, is introduced. The methods are illustrated on synthetic datasets and applied to uncertainty quantification data extracted from the computational chemistry literature.

show abstract

Section: Discussionmentioning

confidence: 99%

“…A recent article by Zheng et al 47 provides formation enthalpies and their uncertainties for two data-driven methods, AIQM1 and ANI-1ccx. The uncertainties are obtained by a query by committee (QbC) strategy, 48 and taken as the standard deviation of the results for an ensemble of neural networks (NN).…”

Section: Zhen2022mentioning

confidence: 99%

Prediction uncertainty validation for computational chemists

Pernot

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Similarly, the ANN was again proven to be capable of predicting semiempirical quantum chemical properties . In one of the most recent studies, a deep-learning ANI-1ccx method was proposed to compute enthalpies of formation of molecules to near chemical accuracy without training directly on experimental values, with time complexity O ( n ) . We are attempting to approach the same level of accuracy and comparable computational cost with another method.…”

Section: Introductionmentioning

confidence: 99%

“…10 In one of the most recent studies, a deep-learning ANI-1ccx method was proposed to compute enthalpies of formation of molecules to near chemical accuracy without training directly on experimental values, with time complexity O(n). 11 We are attempting to approach the same level of accuracy and comparable computational cost with another method. What if we introduce quantum mechanics in between?…”

Section: ■ Introductionmentioning

confidence: 99%

Predicting Experimental Heats of Formation via Deep Learning with Limited Experimental Data

Yang

Chiu

et al. 2022

J. Phys. Chem. A

View full text Add to dashboard Cite

When it comes to predicting experimental values of molecular properties with deep learning, the key problem is the lack of sufficient experimental data for training. We propose a method that consists of pretraining a graph neural network that aims to reproduce first-principles quantum mechanical results, followed by fine-tuning of a fully connected neural network against experimental results. The combined pretraining and fine-tuning model is expected to yield molecular properties close to experimental accuracy. This is made possible because first-principles quantum mechanical methods are often qualitatively correct or semiquantitatively accurate; thus, a calibration of the calculation results against high-precision but limited experiment data can improve accuracy greatly. Moreover, the method is highly efficient, as first-principles quantum mechanical calculation is bypassed. To demonstrate this, we apply the combined model to determine the experimental heats of formation of organic molecules made of H, C, O, N, or F atoms (up to 30 atoms), where mere 405 experimental data are used. The overall mean absolute error is 1.8 kcal/mol for these molecules.

show abstract

“…To give an instance, CASSCF can be made more user-friendly with an automated active-space selection feature using an artificial neural network (ANN), and therefore, it can be used for routine analysis. 18 ML approach has also been efficient in predicting enthalpies of formation, 19 estimation of electronic structure properties of transition metal complexes etc. 20 The ML approach has been applied to choose important configurations.…”

Section: Introductionmentioning

confidence: 99%

Efficient machine learning configuration interaction for bond breaking problems

Rano

Ghosh

2022

Preprint

View full text Add to dashboard Cite

Machine learning-assisted configuration interaction (MLCI) has been shown earlier as a promising method in determining the electronic structure of the model and molecular Hamiltonians. In the MLCI approach to molecular Hamiltonians, it has been noticed that prediction is strongly dependent on the connectedness of the training and validation spaces. In this work, we have tested three different models with different output parameters (abs-MLCI, transformed-MLCI, and log-MLCI) to verify the robustness of training these models. We define robustness as the extent of error in prediction even when the spaces (training and validation) are nected. We notice that the log-MLCI model is best suited to this approach and is, therefore, a powerful model for accurate one-shot variational energies. This is tested not confor chemical bond breaking in water, carbon monoxide, nitrogen, and dicarbon molecules.

show abstract

Toward Chemical Accuracy in Predicting Enthalpies of Formation with General-Purpose Data-Driven Methods

Cited by 27 publications

References 76 publications

Prediction uncertainty validation for computational chemists

Prediction uncertainty validation for computational chemists

Predicting Experimental Heats of Formation via Deep Learning with Limited Experimental Data

Efficient machine learning configuration interaction for bond breaking problems

Contact Info

Product

Resources

About