Adapting hybrid density functionals with machine learning

Science Advances, Journal Year: 2025, Volume and Issue: 11(5)

Published: Jan. 31, 2025

Exact exchange contributions significantly affect electronic states, influencing covalent bond formation and breaking. Hybrid density functional approximations, which average exact admixtures empirically, have achieved success but fall short of high-level quantum chemistry accuracy due to delocalization errors. We propose adaptive hybrid functionals, generating optimal admixture ratios on the fly using data-efficient machine learning models with negligible overhead. The Perdew-Burke-Ernzerhof (aPBE0) improves energetics, electron densities, HOMO-LUMO gaps in QM9, QM7b, GMTKN55 benchmark datasets. A model uncertainty-based constraint reduces method smoothly PBE0 extrapolative regimes, ensuring general applicability limited training. By tuning fractions for different spin aPBE0 effectively addresses gap problem open-shell systems such as carbenes. also present a revised QM9 (revQM9) dataset more accurate properties, including stronger binding, larger bandgaps, localized dipole moments.

Language: Английский

---

An active representation learning method for reaction yield prediction with small-scale data

Communications Chemistry, Journal Year: 2025, Volume and Issue: 8(1)

Published: Feb. 10, 2025

Reaction optimization plays an essential role in chemical research and industrial production. To explore a large reaction system, practical issue is how to reduce the heavy experimental load for finding high-yield conditions. In this paper, we present efficient machine learning tool called "RS-Coreset", where key idea take advantage of deep representation techniques guide interactive procedure representing full space. Our proposed only uses small-scale data, say 2.5% 5% instances, predict yields We validate performance on three public datasets achieve state-of-the-art results. Moreover, apply assist realistic exploration Lewis base-boryl radicals enabled dechlorinative coupling reactions our lab. The can help us effectively even discover several feasible combinations that were overlooked previous articles. Optimization systems crucial production, but represents significant challenge given required find optimal high-yielding Here, authors introduce tool, RS-Coreset, they space, enabling yield prediction with data.

Language: Английский

---

Kernel regression methods for prediction of materials properties: Recent developments

Chemical Physics Reviews, Journal Year: 2025, Volume and Issue: 6(1)

Published: Feb. 13, 2025

Machine learning (ML) is increasingly used in chemical physics and materials science. One major area of thrust machine properties molecules solid from descriptors composition structure. Recently, kernel regression methods various flavors—such as ridge regression, Gaussian process support vector machine—have attracted attention such applications. Kernel allow benefiting simultaneously the advantages linear regressions superior expressive power nonlinear kernels. In many applications, are high-dimensional feature spaces, where sampling with training data bound to be sparse effects specific spaces significantly affect performance method. We review recent applications kernel-based for prediction structure related purposes. discuss methodological aspects including choices kernels appropriate different dimensionality, ways balance reliability model data. also regression-based hybrid ML approaches.

Language: Английский

---

Adaptive Representation of Molecules and Materials in Bayesian Optimization

Chemical Science, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

Feature Adaptive Bayesian Optimization (FABO) enhances molecular and materials discovery by dynamically selecting optimal feature representations during optimization, outperforming fixed representations.

Language: Английский

---

Design and Optimization of a shared synthetic route for multiple active pharmaceutical ingredients through combined Computer Aided Retrosynthesis and flow chemistry

Process Safety and Environmental Protection, Journal Year: 2025, Volume and Issue: unknown

Published: March 1, 2025

Language: Английский

---

Out-of-Distribution Material Property Prediction Using Adversarial Learning

The Journal of Physical Chemistry C, Journal Year: 2025, Volume and Issue: unknown

Published: March 25, 2025

The accurate prediction of material properties is crucial in a wide range scientific and engineering disciplines. Machine learning (ML) has advanced the state art this field, enabling scientists to discover novel materials design with specific desired properties. However, one major challenge that persists property generalization models out-of-distribution (OOD) samples, i.e., samples differ significantly from those encountered during training. In real-world discovery, OOD scenarios often arise when applying ML predict additional within newly explored region originating few experimental samples. paper, we explore application advancements approaches enhance robustness reliability models. We propose apply Crystal Adversarial Learning (CAL) algorithm for prediction, which generates synthetic data training guide toward high uncertainty. further an adversarial learning-based targeted approach make model adapt particular set, as alternative traditional fine-tuning. Our experiments suggest our CAL can be effective limited commonly occur science. work provides important step improved highlights areas require exploration refinement.

Language: Английский

---

Miniaturized Supercritical Fluid Chromatography Coupled with Ion Mobility Spectrometry: A Chip-Based Platform for Rapid Chiral and Complex Mixture Analysis

Analytical Chemistry, Journal Year: 2025, Volume and Issue: unknown

Published: April 5, 2025

This study presents the first coupling of miniaturized chip-based supercritical fluid chromatography (SFC) with ion mobility spectrometry (IMS) enabling rapid two-dimensional analysis moderately polar compounds. For time, ionization and analyte transfer at SFC-IMS interface are achieved solely through eluent decompression in conjunction a shifted electric IMS inlet potential. straightforward approach significantly reduces instrumentation complexity size, promoting system compactness robustness. The integration SFC enables high-speed separations complex samples, drastically reducing time while utilizing detector capable delivering structural information acquisition rate low cost. Evaluation as demonstrated chiral separation Tröger's base revealed exceptional repeatability sensitivity. Short columns high flow rates resulted record-speed just six seconds. was successfully used to analyze mixture containing five isomers, including naloxone 6-monoacetylmorphine, 30 s.

Language: Английский

---

Developing ChemDFM as a large language foundation model for chemistry

Cell Reports Physical Science, Journal Year: 2025, Volume and Issue: 6(4), P. 102523 - 102523

Published: April 1, 2025

Language: Английский

---

Optimal transport for generating transition states in chemical reactions

Nature Machine Intelligence, Journal Year: 2025, Volume and Issue: 7(4), P. 615 - 626

Published: April 23, 2025

Language: Английский

---

A FAIR comparison of activated carbon, biochar, cyclodextrins, polymers, resins, and metal organic frameworks for the adsorption of per- and polyfluorinated substances

Chemical Engineering Journal, Journal Year: 2024, Volume and Issue: 498, P. 155456 - 155456

Published: Sept. 3, 2024

Per- and polyfluorinated substances (PFAS) have complex sorption behaviors, complicating removal from water selection of suitable adsorbents. We evaluated adsorption 44 PFAS across four adsorbent groups: activated carbon biochar (AC BC), cyclodextrin-based adsorbents (cyclodextrins), polymer-based resins, inorganic metal organic frameworks (MOFs). analyzed over 500 coefficients (Kd) literature, calculated at aqueous equilibrium concentration 1 ± 0.3 µg/L under comparable experimental conditions. On average, Kd AC BC exceeded 107 L/kg for with C-F bonds > 7, unlike other < 107. This trend holds 4. Cyclodextrins, resins outperform ≤ For BC, follows the order PFOS>PFOA>PFBS>PFBA, increasing point zero charge. as well cyclodextrin, values were related to hydrophobicity steric properties. Additionally, was influenced by head group type, non-fluorinated atoms, presence oxygen and/or chlorine in PFAS. No clear relationship found Adsorption prediction using a Random Forest Regressor literature data feasible but not Cyclodextrins removing varying mobilities water, whereas are superior low mobility To support further use all code used freely available, following FAIR principles.

Language: Английский

---