Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
CPP: Fachverband Chemische Physik und Polymerphysik
CPP 7: Emerging Topics in Chemical and Polymer Physics, New Instruments and Methods I
CPP 7.4: Vortrag
Montag, 9. März 2026, 12:15–12:30, ZEU/0255
MolecuTas: an ML platform for refining quantum properties and bioactivity of complex molecules — •Álvaro Vallejo Bay1, Jannis Krüger2, Thomas Hellweg2, Gerardo Prieto1, and Vicente Domínguez Arca2,3 — 1Applied Physics, University of Santiago de Compostela, Spain — 2Physical and Biophysical Chemistry, Bielefeld University, Germany — 3Biosystems and Bioprocesses Engineering, IIM-CSIC, Spain
This work introduces MolecuTas, a neural architecture designed for the ultrafast prediction of atomistic and molecular properties with accuracy approaching density functional theory, yet at negligible computational cost. The project tackles core challenges in computational chemistry, particularly the size dependence of predictive models and the loss of structural detail in conventional representations. To overcome these limitations, we develop a custom data extraction and molecular fragmentation pipeline based on large quantum datasets, ensuring the preservation of essential information such as connectivity, symmetries, chemical environments, and electronic features.
Building on this foundation, we propose a family of advanced Graph Neural Networks that explicitly integrate physicochemical principles and disentangle local from global information, enabling robust generalization across diverse molecular systems. Ultimately, MolecuTas aims to provide a reliable tool for molecular dynamics, enabling precise and fast prediction of partial charges and other key descriptors required for accurate simulations and next-generation chemical design.
Keywords: MolecuTas; Quantum Property Prediction; Graph Neural Networks; Machine Learning in Material Science; Drug Discovery