Application of Machine Learning in Predicting Molecular Properties
Scholarly Review Journal · 2022
DOI: 10.70121/001c.121662 ↗Kurz gefasst
Fachartikel in der Zeitschrift Scholarly Review Journal, 2022. Verfasst von Edward Sun. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
As developing new compounds and determining their properties is expensive and possibly dangerous, it becomes necessary to develop a model to predict the molecular properties without having to synthesize and test it experimentally. Two systematic ways to represent a compound are through the schematic diagram of the molecular structure and the simplified molecular-input line-entry system (SMILES). In this study, these representations were used to train two neural network models, a convolutional neural network (CNN) and a recurrent neural network (RNN), respectively, in order to predict the compounds’ melting points. By representing the compounds as an image of the structure, the CNN was unsuccessful at fitting the given data, appearing to remain constant at around the average melting point of the given data. However, by representing the compounds as a systematically generated string of text, the RNN was successful at fitting the data, with an overall trend similar to the actual trend and a lower mean absolute error. However, the SMILES data used for the RNN does not contain orientation information, unlike the structure diagram data. For future studies, it may be possible to combine the two representations to reach a more accurate prediction model.
Zitieren
APA
Sun, E. (2022). Application of Machine Learning in Predicting Molecular Properties. Scholarly Review Journal. https://doi.org/10.70121/001c.121662
Vancouver
Sun E. Application of Machine Learning in Predicting Molecular Properties. Scholarly Review Journal. 2022. doi:10.70121/001c.121662
BibTeX
@article{Sun2022,
author = {Sun, Edward},
title = {Application of Machine Learning in Predicting Molecular Properties},
journal = {Scholarly Review Journal},
year = {2022},
doi = {10.70121/001c.121662}
}
EUVIMED