The Alshami Research Group’s commitment to innovation extends beyond the laboratory, as they integrate Artificial Intelligence (AI) and Machine Learning (ML) methods into their research. The group has been engaged in three distinct projects utilizing ML methods.
In the first project, two different ML methods were developed to accurately predict the solubility of various species using data from over 8400 compounds. Molecular descriptors, commonly used in previous studies, and Morgan fingerprints, circular-based hashes of molecules’ structures, were applied to generate water solubility estimates. The significance of this study lies in the practical utility of the developed fingerprint model, which can assist experts in investigating the impact of different functional groups on solubility predictions. This has important implications for drug discovery and other related applications.
Link to the study: https://link.springer.com/article/10.1186/s13321-023-00752-6
In another research endeavor, the group explored the potential of using machine learning (ML) to graft an RO membrane’s polyamide (PA) surface, aiming to increase water permeability and overcome the permeability/selectivity tradeoff limitations. Moieties with positive and negative contributions toward water permeability were identified using Shapley-Additive-explanations (SHAP) analysis as an explainable artificial intelligence (XAI) method. By improving the subunits of the PA’s structure with positive Shapley values, the polyamide RO membrane layer of a commercial membrane, Dupont XLE, experienced a substantial increase in water permeability. A path from A to Z that demonstrates incorporating ML algorithms into membrane fabrication processes can be found in our published study.
Link to the study: https://www.sciencedirect.com/science/article/pii/S2772421222000204
A machine learning (ML) algorithm was also developed to predict scale formation in produced water samples. The intention is to deploy this model on a static website without needing a server. This approach shifts computational requirements onto website visitors, eliminating the necessity for installation and other computational software licenses.
To access the model, click the “SCALE PREDICTOR TRAINED MODEL” tab on the home page.