A Novel Crystal Structure Prediction Using Hybrid Method DOI Open Access

Heren Chellam G.

Tuijin Jishu/Journal of Propulsion Technology, Journal Year: 2023, Volume and Issue: 44(4), P. 5356 - 5365

Published: Nov. 15, 2023

Chemical compositions are used to predict the crystal structure in solid state of new materials. To finding crystalline arrangements materials for major unsolved problems science their chemical compositions. Crystal prediction is one foremost methods discovering In this paper, we propose a deep and machine learning model approach classification structure. The more than 5000 dataset were previous work, various models predicting fuse neural network algorithm. trained tested prediction. evaluate with high accuracy. Our approach, ANB-NET(AlextNet Naive Bayes) classifier get best accuracy time complexity less other model.

Language: Английский

Recent Advances in the Application of Machine Learning to Crystal Behavior and Crystallization Process Control DOI
Meijin Lu, Silin Rao, Hong Yue

et al.

Crystal Growth & Design, Journal Year: 2024, Volume and Issue: 24(12), P. 5374 - 5396

Published: June 6, 2024

Crystals are integral to a variety of industrial applications, such as the development pharmaceuticals and advancements in material science. To anticipate crystal behavior pinpoint effective crystallization techniques, thorough investigation structures, properties, associated processes is essential. However, conventional methods like experimental procedures quantum mechanics calculations, while crucial, can be expensive time-consuming. In response, machine learning has risen an alternative, complementing traditional approaches based on classical force fields. recent years, deployment realm yielded notable progress. This review offers concise overview application techniques crystallization, focusing past five years. Our analysis literature indicates that accelerated prediction structures by streamlining generation evaluation structures. Additionally, it facilitated key properties solubility, melting point, habit. The further explores role refining control optimization processes, highlighting restrictions algorithms sensing technologies. advantages end-to-end processing for enhancing accuracy predictions combination data-driven with mechanism-based models robustness also considered. summary, this provides insights into current state field intelligent suggests pathways future research development.

Language: Английский

Citations

8
Global machine learning potentials for molecular crystals DOI Creative Commons
Ivan Žugec, R. Matthias Geilhufe, Ivor Lončarić

et al.

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 160(15)

Published: April 16, 2024

Molecular crystals are difficult to model with accurate first-principles methods due large unit cells. On the other hand, modeling is required as polymorphs often differ by only 1 kJ/mol. Machine learning interatomic potentials promise provide accuracy of baseline a cost lower orders magnitude. Using existing databases density functional theory calculations for molecular and molecules, we train global machine potentials, usable any crystal. We test performance on experimental benchmarks show that they perform better than classical force fields and, in some cases, comparable calculations.

Language: Английский

Citations

3
Impact of heteroatoms and chemical functionalisation on crystal structure and carrier mobility of organic semiconductors DOI Creative Commons

Sebastian Hutsch,

Frank Ortmann

npj Computational Materials, Journal Year: 2024, Volume and Issue: 10(1)

Published: Sept. 4, 2024

Language: Английский

Citations

2
Prognostic prediction models for postoperative patients with stage I to III colorectal cancer based on machine learning DOI Open Access

Xiaolin Ji,

Shuo Xu, Xiaoyu Li

et al.

World Journal of Gastrointestinal Oncology, Journal Year: 2024, Volume and Issue: 16(12), P. 4597 - 4613

Published: Nov. 8, 2024

BACKGROUND Colorectal cancer (CRC) is characterized by high heterogeneity, aggressiveness, and morbidity mortality rates. With machine learning (ML) algorithms, patient, tumor, treatment features can be used to develop validate models for predicting survival. In addition, important variables screened different applications provided that could serve as vital references when making clinical decisions potentially improving patient outcomes in settings. AIM To construct prognostic prediction screen patients with stage I III CRC. METHODS More than 1000 postoperative CRC were grouped according survival time (with cutoff values of 3 years 5 years) assigned training testing cohorts (7:3). For each 3-category time, predictions made 4 ML algorithms (all-variable variable-only datasets), which was validated via 5-fold cross-validation bootstrap validation. Important multivariable regression methods. Model performance evaluated compared before after variable screening the area under curve (AUC). SHapley Additive exPlanations (SHAP) further demonstrated impact on model decision-making. Nomograms constructed practical application. RESULTS Our performed well; parameter identification consistent, effective. The highest pre- postscreening AUCs 95% confidence intervals set 0.87 (0.81-0.92) 0.89 (0.84-0.93) overall survival, 0.75 (0.69-0.82) 0.73 (0.64-0.81) disease-free 0.95 (0.88-1.00) 0.88 (0.75-0.97) recurrence-free 0.76 (0.47-0.95) 0.80 (0.53-0.94) distant metastasis-free Repeated validation both datasets. SHAP consistent clinicopathological characteristics tumors. nomograms created. CONCLUSION We a comprehensive, high-accuracy, variable-based architecture times. This reference managing patients.

Language: Английский

Citations

1
Predicting Miscibility in Binary Compounds: A Machine Learning and Genetic Algorithm Study DOI

Chiwen Feng,

Yanwei Liang,

Jiaying Sun

et al.

Physical Chemistry Chemical Physics, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 13, 2024

This study used atomic-level data and machine learning to predict the miscibility of binary systems, analyzed key factors affecting miscibility, discovered three new thermodynamically stable phases using a genetic algorithm.

Language: Английский

Citations

1
An open science grid implementation of the steady state genetic algorithm for crystal structure prediction DOI
Kristal N. Varela, Gabriel I. Pagola,

Albert M. Lund

et al.

Journal of Computational Science, Journal Year: 2024, Volume and Issue: 82, P. 102415 - 102415

Published: Aug. 14, 2024

Language: Английский

Citations

0
Topological relations between crystal structures: a route to predicting inorganic materials DOI
Natalia A. Kabanova,

Ekaterina A. Grishina,

Vladislav T. Osipov

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 6, 2024

Abstract We review topological approaches to the analysis of crystal structures intermetallic compounds and searching for structural relations between them as their underlying atomic nets. introduce concept skeletal net find simplest system interatomic contacts in compounds, which supports three-periodic architecture. Using observed we have revealed binary MeX (Me = Re, Ti or Rh; X B, C, N Si) found a key role body-centered cubic hierarchy. explored configuration space corresponding crystalline systems by generating all possible ‘subnet-supernet’ transformations, optimized resulting motifs with DFT methods new phase RhB be stable above 22 GPa. discuss representations prediction chemical substances.

Language: Английский

Citations

0
A Novel Crystal Structure Prediction Using Hybrid Method DOI Open Access

Heren Chellam G.

Tuijin Jishu/Journal of Propulsion Technology, Journal Year: 2023, Volume and Issue: 44(4), P. 5356 - 5365

Published: Nov. 15, 2023

Chemical compositions are used to predict the crystal structure in solid state of new materials. To finding crystalline arrangements materials for major unsolved problems science their chemical compositions. Crystal prediction is one foremost methods discovering In this paper, we propose a deep and machine learning model approach classification structure. The more than 5000 dataset were previous work, various models predicting fuse neural network algorithm. trained tested prediction. evaluate with high accuracy. Our approach, ANB-NET(AlextNet Naive Bayes) classifier get best accuracy time complexity less other model.

Language: Английский

Citations

0