Graph neural network for descriptor-free CO2 solubility prediction in aqueous amines

Name: Apri Wahyudi

Organization : Chulalongkorn University. The Petroleum and Petrochemical College

Name: Natthapong Sueviriyapan

Organization : Chulalongkorn University. The Petroleum and Petrochemical College

Email : natthapong.su@chula.ac.th

Name: Uthaiporn Suriyapraphadilok

Organization : Chulalongkorn University. The Petroleum and Petrochemical College

keyword: Amine solutions

LCSH: Carbon dioxide

LCSH: Amine oxidase

LCSH: Carbon sequestration

LCSH: Machine learning

Abstract: The development of an effective carbon capture unit via chemical absorption depends on the knowledge of CO₂ solubility in amine solutions. Conventional modeling methods lack flexibility in solvent representation by relying on predetermined molecular descriptors, such as the COSMO σ-profile, which requires computationally expensive quantum chemistry calculations. This study presents a graph neural network (GNN) for learning molecular representations directly from SMILES notations, thereby removing the reliance on predefined descriptors. Our approach offers flexible and reliable solubility prediction for 61 species of single amine systems over wide ranges of concentrations, temperatures, and CO₂ partial pressures, achieving equivalent accuracy for the training and validation sets compared to descriptor-based approaches (R² > 0.97, RMSE < 0.07 mol CO₂/mol amine). Under industrially relevant conditions, the model effectively predicted CO₂ solubility for different amines, including well-known amines such as monoethanolamine (MEA), piperazine (PZ), and 2-(ethylamino)ethanol (EMEA). Our proposed method allows for the rapid screening of new amine structures without relying on feature engineering. This enables inverse molecular design—identifying ideal amine structures for specific operating conditions—potentially accelerating the development of better solvents. This study demonstrates how graph-based machine learning can overcome the constraints of conventional descriptor-based models, offering a more flexible approach to rational solvent design for carbon capture applications.

King Mongkut's University of Technology North Bangkok. Central Library

Address: BANGKOK

Email: library@kmutnb.ac.th

Created: 2025

Modified: 2025-07-25

Issued: 2025-07-25

บทความ/Article

application/pdf

BibliograpyCitation : In Thai Institute of Chemical Engineering and Applied Chemistry and Khon Kaen University. Faculty of Engineering. The 34th Thai Institute of Chemical Engineering and Applied Chemistry International Conference (TIChE 2025) (TIChE 2025 TIChE-PE-03). Bangkok : Thai Institute of Chemical Engineering and Applied Chemistry, 2025

eng

©copyrights King Mongkut's University of Technology North Bangkok

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	TIChE 2025 TIChE-PE-03.pdf	2 MB

ใช้เวลา

0.02824 วินาที