Automatic Prediction of Band Gaps of Inorganic Materials Using a Gradient Boosted and Statistical Feature Selection Workflow

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: 64(4), P. 1187 - 1200

Published: Feb. 6, 2024

Machine learning (ML) methods can train a model to predict material properties by exploiting patterns in materials databases that arise from structure-property relationships. However, the importance of ML-based feature analysis and selection is often neglected when creating such models. Such are especially important dealing with multifidelity data because they afford complex space. This work shows how gradient-boosted statistical feature-selection workflow be used predictive models classify their metallicity band gap against experimental measurements, as well computational derived electronic-structure calculations. These fine-tuned via Bayesian optimization, using solely features chemical compositions data. We test these experimental, computational, combination find modeling option reduce number required model. The performance our benchmarked state-of-the-art algorithms, results which demonstrate approach either comparable or superior them. classification realized an accuracy score 0.943, macro-averaged F1-score 0.940, area under curve receiver operating characteristic 0.985, average precision 0.977, while regression achieved mean absolute error 0.246, root-mean squared 0.402,

Language: Английский

---

Automatic Prediction of Peak Optical Absorption Wavelengths in Molecules Using Convolutional Neural Networks

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: 64(5), P. 1486 - 1501

Published: Feb. 29, 2024

Molecular design depends heavily on optical properties for applications such as solar cells and polymer-based batteries. Accurate prediction of these is essential, multiple predictive methods exist, from ab initio to data-driven techniques. Although theoretical methods, time-dependent density functional theory (TD-DFT) calculations, have well-established physical relevance are among the most popular in computational physics chemistry, they exhibit errors that inherent their approximate nature. These high-throughput electronic structure calculations also incur a substantial cost. With emergence big-data initiatives, cost-effective, gained traction, although usability highly contingent degree data quality sparsity. In this study, we present workflow employs deep residual convolutional neural networks (DR-CNN) gradient boosting feature selection predict peak absorption wavelengths (λmax) exclusively SMILES representations dye molecules solvents; one would normally measure λmax using UV–vis spectroscopy. We use multifidelity modeling approach, integrating 34,893 DFT 26,395 experimentally derived data, deliver more accurate predictions via Bayesian-optimized machine. Our approach benchmarked against state art reported scientific literature; results demonstrate learnt DR-CNN integrated with other machine learning can accelerate specific characteristics.

Language: Английский

---

Introduction to Machine Learning for Predictive Modeling I

Challenges and advances in computational chemistry and physics, Journal Year: 2025, Volume and Issue: unknown, P. 3 - 30

Published: Jan. 1, 2025

Language: Английский

---

Spatial-Temporal Dynamics in Country-Level Sustainable Energy Performance Using Ensemble Learning and Analytic Hierarchy Process

Journal of Cleaner Production, Journal Year: 2025, Volume and Issue: unknown, P. 145497 - 145497

Published: April 1, 2025

Language: Английский

---

Machine-Learning Prediction of Curie Temperature from Chemical Compositions of Ferromagnetic Materials

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: 64(16), P. 6388 - 6409

Published: Aug. 7, 2024

Room-temperature ferromagnets are high-value targets for discovery given the ease by which they could be embedded within magnetic devices. However, multitude of potential interactions among ions and their surrounding environments renders prediction thermally stable properties challenging. Therefore, it is vital to explore methods that can effectively screen candidates expedite novel ferromagnetic materials highly intricate feature spaces. To this end, we machine-learning (ML) as a means predict Curie temperature (Tc) discerning patterns databases. This study emphasizes importance analysis selection in ML modeling demonstrates efficacy our gradient-boosted statistical feature-selection workflow training predictive models. The models fine-tuned through Bayesian optimization, using features derived solely from chemical compositions data, before model predictions evaluated against literature values. We have collated ca. 35,000 Tc values performance benchmarked state-of-the-art algorithms, results demonstrate methodology superior majority alternative methods. In 10-fold cross-validation, regression realized an R2 (0.92 ± 0.01), MAE (40.8 1.9) K, RMSE (80.0 5.0) K. utility case studies forecast rare-earth intermetallic compounds generate phase diagrams various systems. These highlight systematic approach enhancing both capability interpretability models, while being devoid human bias. They advantages such over mere reliance on algorithmic complexity black-box treatment ML-based domain computational science.

Language: Английский

---

Machine-Learning Predictions of Critical Temperatures from Chemical Compositions of Superconductors

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 17, 2024

In the quest for advanced superconducting materials, accurate prediction of critical temperatures (

Language: Английский

---

Predictive Modeling of High-Entropy Alloys and Amorphous Metallic Alloys Using Machine Learning

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 1, 2024

High entropy alloys and amorphous metallic represent two distinct classes of advanced alloy materials, each with unique structural characteristics. Their emergence has garnered considerable interest across the materials science engineering communities, driven by their promising properties, including exceptional strength. However, extensive compositional diversity poses substantial challenges for systematic exploration, as traditional experimental approaches high-throughput calculations struggle to efficiently navigate this vast space. While recent development in data-driven discovery could potentially help, such efforts are hindered scarcity comprehensive data lack robust predictive tools that can effectively link composition specific properties. To address these challenges, we have deployed a machine-learning-based workflow feature selection statistical analysis afford models accelerate optimization materials. Our methodology is validated through case studies: (i) regression bulk modulus, (ii) classification based on glass-forming ability. The Bayesian-optimized model trained prediction modulus achieved an

Language: Английский

---

Machine Learning Approaches for Predicting Power Conversion Efficiency in Organic Solar Cells: A Comprehensive Review

Solar RRL, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 9, 2024

Organic solar cells (OSCs), renowned for their lightweight, cost efficiency, and adaptability nature, stand out as a promising option developing renewable energy. Improving the power conversion efficiency (PCE) of OSCs is essential, researchers are delving into novel materials to achieve this. Traditional approaches often laborious costly, highlighting need predictive modeling. Machine learning (ML), especially via quantitative structure–property relationship (QSPR) models, streamlining material development, with goal exceed 20% PCE. In this review, application ML in explored, recent studies utilizing PCE prediction reviewed, encompassing empirical functions, algorithms, self‐devised frameworks, combination automated experimental technologies. First, benefits predicting addressed. Second, development high‐efficiency models both fullerene nonfullerene acceptors delved into. The impact various algorithm on then assessed, taking account construction models. Moreover, quality databases selection descriptors considered. Databases based further categorized. Finally, prospects future proposed.

Language: Английский

---

Automatic Prediction of Molecular Properties Using Substructure Vector Embeddings within a Feature Selection Workflow

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 23, 2024

Machine learning (ML) methods provide a pathway to accurately predict molecular properties, leveraging patterns derived from structure–property relationships within materials databases. This approach holds significant importance in drug discovery and design, where the rapid, efficient screening of molecules can accelerate development new pharmaceuticals chemical for highly specialized target application. Unsupervised self-supervised applied graph-based or geometric models have garnered considerable traction. More recently, transformer-based language emerged as powerful tools. Nevertheless, their application entails computational resources, owing need an extensive pretraining process on vast corpus unlabeled data sets. To this end, we present semisupervised strategy that harnesses substructure vector embeddings conjunction with ML-based feature selection workflow various properties. We evaluate efficacy our modeling methodology across diverse range sets, encompassing both regression classification tasks. Our findings demonstrate superior performance compared most existing state-of-the-art algorithms, while offering advantages terms balancing model accuracy requirements. Moreover, provides deeper insights into interactions are essential interpretability. A case study is conducted lipophilicity molecules, exemplifying robustness strategy. The result underscores meticulous analysis over mere reliance predictive high degree algorithmic complexity.

Language: Английский

---

Negative Poisson's ratio of sulfides dominated by strong intralayer electron repulsion

Physical Chemistry Chemical Physics, Journal Year: 2024, Volume and Issue: 26(31), P. 20852 - 20863

Published: Jan. 1, 2024

Geometrical variations in a particular structure or other mechanical factors are often cited as the cause of negative Poisson's ratio (NPR).

Language: Английский

---