Maximum diversification of data is a central theme in building generalized and accurate machine learning (ML) models. In chemistry, ML has been used to develop models for predicting molecular properties, for example quantum mechanics (QM) calculated potential energy surfaces and atomic charge models. The ANI-1x and ANI-1ccx ML-based general-purpose potentials for organic molecules were developed through active learning; an automated data diversification process. Here, we describe the ANI-1x and ANI-1ccx data sets. To demonstrate data diversity, we visualize it with a dimensionality reduction scheme, and contrast against existing data sets. The ANI-1x data set contains multiple QM properties from 5 M density functional theory calculations, while the ANI-1ccx data set contains 500 k data points obtained with an accurate CCSD(T)/CBS extrapolation. Approximately 14 million CPU core-hours were expended to generate this data. Multiple QM calculated properties for the chemical elements C, H, N, and O are provided: energies, atomic forces, multipole moments, atomic charges, etc. We provide this data to the community to aid research and development of ML models for chemistry.more » « less
- NSF-PAR ID:
- Publisher / Repository:
- Nature Publishing Group
- Date Published:
- Journal Name:
- Scientific Data
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
Modern semiempirical electronic structure methods have considerable promise in drug discovery as universal “force fields” that can reliably model biological and drug-like molecules, including alternative tautomers and protonation states. Herein, we compare the performance of several neglect of diatomic differential overlap-based semiempirical (MNDO/d, AM1, PM6, PM6-D3H4X, PM7, and ODM2), density-functional tight-binding based (DFTB3, DFTB/ChIMES, GFN1-xTB, and GFN2-xTB) models with pure machine learning potentials (ANI-1x and ANI-2x) and hybrid quantum mechanical/machine learning potentials (AIQM1 and QD π) for a wide range of data computed at a consistent ωB97X/6-31G* level of theory (as in the ANI-1x database). This data includes conformational energies, intermolecular interactions, tautomers, and protonation states. Additional comparisons are made to a set of natural and synthetic nucleic acids from the artificially expanded genetic information system that has important implications for the design of new biotechnology and therapeutics. Finally, we examine the acid/base chemistry relevant for RNA cleavage reactions catalyzed by small nucleolytic ribozymes, DNAzymes, and ribonucleases. Overall, the hybrid quantum mechanical/machine learning potentials appear to be the most robust for these datasets, and the recently developed QD π model performs exceptionally well, having especially high accuracy for tautomers and protonation states relevant to drug discovery.more » « less
AQME, automated quantum mechanical environments, is a free and open‐source Python package for the rapid deployment of automated workflows using cheminformatics and quantum chemistry. AQME workflows integrate tasks performed across multiple computational chemistry packages and data formats, preserving all computational protocols, data, and metadata for machine and human users to access and reuse. AQME has a modular structure of independent modules that can be implemented in any sequence, allowing the users to use all or only the desired parts of the program. The code has been developed for researchers with basic familiarity with the Python programming language. The CSEARCH module interfaces to molecular mechanics and semi‐empirical QM (SQM) conformer generation tools (e.g., RDKit and Conformer–Rotamer Ensemble Sampling Tool, CREST) starting from various initial structure formats. The CMIN module enables geometry refinement with SQM and neural network potentials, such as ANI. The QPREP module interfaces with multiple QM programs, such as Gaussian, ORCA, and PySCF. The QCORR module processes QM results, storing structural, energetic, and property data while also enabling automated error handling (i.e., convergence errors, wrong number of imaginary frequencies, isomerization, etc.) and job resubmission. The QDESCP module provides easy access to QM ensemble‐averaged molecular descriptors and computed properties, such as NMR spectra. Overall, AQME provides automated, transparent, and reproducible workflows to produce, analyze and archive computational chemistry results. SMILES inputs can be used, and many aspects of tedious human manipulation can be avoided. Installation and execution on Windows, macOS, and Linux platforms have been tested, and the code has been developed to support access through Jupyter Notebooks, the command line, and job submission (e.g., Slurm) scripts. Examples of pre‐configured workflows are available in various formats, and hands‐on video tutorials illustrate their use.
This article is categorized under:
Data Science > Chemoinformatics
Data Science > Computer Algorithms and Programming
Software > Quantum Chemistry
Computational modeling of chemical and biological systems at atomic resolution is a crucial tool in the chemist’s toolset. The use of computer simulations requires a balance between cost and accuracy: quantum-mechanical methods provide high accuracy but are computationally expensive and scale poorly to large systems, while classical force fields are cheap and scalable, but lack transferability to new systems. Machine learning can be used to achieve the best of both approaches. Here we train a general-purpose neural network potential (ANI-1ccx) that approaches CCSD(T)/CBS accuracy on benchmarks for reaction thermochemistry, isomerization, and drug-like molecular torsions. This is achieved by training a network to DFT data then using transfer learning techniques to retrain on a dataset of gold standard QM calculations (CCSD(T)/CBS) that optimally spans chemical space. The resulting potential is broadly applicable to materials science, biology, and chemistry, and billions of times faster than CCSD(T)/CBS calculations.
As machine learning (ML) has matured, it has opened a new frontier in theoretical and computational chemistry by offering the promise of simultaneous paradigm shifts in accuracy and efficiency. Nowhere is this advance more needed, but also more challenging to achieve, than in the discovery of open‐shell transition metal complexes. Here, localized
dor felectrons exhibit variable bonding that is challenging to capture even with the most computationally demanding methods. Thus, despite great promise, clear obstacles remain in constructing ML models that can supplement or even replace explicit electronic structure calculations. In this article, I outline the recent advances in building ML models in transition metal chemistry, including the ability to approach sub‐kcal/mol accuracy on a range of properties with tailored representations, to discover and enumerate complexes in large chemical spaces, and to reveal opportunities for design through analysis of feature importance. I discuss unique considerations that have been essential to enabling ML in open‐shell transition metal chemistry, including (a) the relationship of data set size/diversity, model complexity, and representation choice, (b) the importance of quantitative assessments of both theory and model domain of applicability, and (c) the need to enable autonomous generation of reliable, large data sets both for ML model training and in active learning or discovery contexts. Finally, I summarize the next steps toward making ML a mainstream tool in the accelerated discovery of transition metal complexes.
This article is categorized under:
Electronic Structure Theory > Density Functional Theory
Software > Molecular Modeling
Computer and Information Science > Chemoinformatics
Machine learning (ML) regression methods are promising tools to develop models predicting the properties of materials by learning from existing databases. However, although ML models are usually good at interpolating data, they often do not offer reliable extrapolations and can violate the laws of physics. Here, to address the limitations of traditional ML, we introduce a “topology-informed ML” paradigm—wherein some features of the network topology (rather than traditional descriptors) are used as fingerprint for ML models—and apply this method to predict the forward (stage I) dissolution rate of a series of silicate glasses. We demonstrate that relying on a topological description of the atomic network (i) increases the accuracy of the predictions, (ii) enhances the simplicity and interpretability of the predictive models, (iii) reduces the need for large training sets, and (iv) improves the ability of the models to extrapolate predictions far from their training sets. As such, topology-informed ML can overcome the limitations facing traditional ML (e.g., accuracy vs. simplicity tradeoff) and offers a promising route to predict the properties of materials in a robust fashion.