Work

A Multimodal Learning Model based on a QSPR approach for the estimation of RON, MON and CN, for any C, H, O hydrocarbons

Year: 2024

Type: article

Abstract: With the increasing demand in alternative fuels and the use of biomass as the feedstock for the production of fuel commodities, a wider range of oxygenated hydrocarbons as fuel additives needs to be considered. Consequently, the development of robust method for the prediction of criterium as Research Octane Number (RON), Motor Octane Number (MON) or Cetane Number (CN), will play a crucial role in the characterization of novel fuels. In this paper we proposed a robust deep-learning model based on a Quantitative Structure−Property Relationship (QSPR) approach for the estimation of RON, MON and CN of any C, H, O molecules. We develop a multimodal learning model from a combination of two types of data and based on an Artificial Neural Network. The Mordred algorithm has been used in order the determine 457 descriptors to characterize any hydrocarbons. These numerical values represent the first type of data considered in the present study. To consider the effect of mesomerism or chirality of molecules the InChIKey notation has been used. This notation, composed by 27 letters, represent the second type of data and are considered as text data. To encode textual variables into numeric data, we have used a Word Embedding method. The predictions of the final model have been tested with success on a large set of experimental data and compared with those of five recent learning models, GNN, ANN, GPR, DLMO and PIGNN, of literature. The methodology of GNN (Graph Neural Networks) model is based on the architecture of the molecule, that of the ANN (Artificial Neural Networks) is based on a limited number of chemical groups, whereas the GPR (Gaussian Process Regression) model is mainly based on Joback groups. The two most recent methodologies DLMO (Deep Learning Mixing Operator) and PIGNN (Physics-Informed Graph Neural Network) are based on more sophisticated algorithms. The full comparison and several tests show a very robust and predictive ability of our newly proposed multimodal learning model. The prediction tool is available via a web page located at http://ehlcathol.eu/.

Source: Fuel

DOI: 10.1016/j.fuel.2024.133438

Authors: Roda Bounaceur, Nicolas Barthélemy, Nicolas Delort, Olivier Herbinet, Frédérique Battin-Leclerc

Institutions: Centre National de la Recherche Scientifique, Laboratoire Réactions et Génie des Procédés, Université de Lorraine

Language: English

Topic: Computational Drug Discovery Methods

Subfield: Computational Theory and Mathematics

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Estimated APC paid: $4,200

Funder: HORIZON EUROPE Reforming and enhancing the European Research and Innovation system