- Home
- Industry Solutions
- Services
- Support
- About Us
Aqueous solubility is crucial to the processes of drug discovery and development. It is an important factor affecting oral absorption and bioavailability of drugs and is considered a relevant parameter in ADMET (absorption, distribution, metabolism, excretion, and toxicity) studies. Many drug development failures have been linked to poor solubility, and increasing the water solubility of bioactive compounds is a significant challenge in medicinal chemistry. In addition, aqueous solubility is also a key determinant of the environmental impact of pollutants and agricultural chemicals. To determine the water solubility compounds, a variety of experimental techniques have been used, such as variations of the shake-flask method and the CheqSol approach. However, experimental methods are difficult, expensive, and time-consuming to determine the aqueous solubility. It is also unrealistic to test thousands or millions of compounds in high throughput screening (HTS). Therefore, the prediction of solubility by in silico approaches is highly valuable. Based on the advanced AI-assisted platform, Creative Biolabs can accurately predict the ADMET properties and help customers accelerate the drug screening process and reduce R&D costs.
Fig.1 Pruned machine learning models to predict aqueous solubility. (Perryman, et al., 2020)
Water solubility is an important physicochemical property of compounds in anticancer drug discovery and development, which affects pharmacokinetic properties and formulations. A free and open-source software program named the Konstanz Information Miner (KNIME) has emerged as one of the most important analytical platforms for innovation, discovering the hidden nature of data, and predicting new features. Through the use of interconnected nodes, KNIME integrates several machine learning and data mining components that can be easily applied to the fields of chemistry, drug design, biology, and the prediction of ADMET features. The flexibility of workflows developed in KNIME to include different tools allows users to read, create, edit, train, and test ML models, greatly facilitating the automation of predictions and applications by any user. Given the limited predictive performance of many published solubility models, some groups have developed innovative QSPR models using new recursive algorithms in machine learning methods for data and variable selection. During the early phases of drug discovery and development, the automatic workflow showed high predictive performance and can offer superior predictions of aqueous solubility. For the prediction of aqueous solubility, some machine learning (ML) methods have been used, including the convolutional and recurrent networks, random forests (RF), support vector machines (SVM), and k-nearest neighbors (k-NN).
Fig.2 Deep learning architectures for aqueous solubility prediction. (Panapitiya, et al., 2022)
A variety of artificial intelligence solubility prediction tools have been developed by utilizing deep learning, regression, and modeling machine learning to facilitate solubility assessment. These tools have achieved outstanding results with high R2 and low RMSE values. However, because different data sets are used, the reported performance can vary considerably even with the same tools. It is necessary to enhance solubility prediction for novel compounds, which can be further achieved through deep learning. Solubility prediction may improve as deep learning progresses. The deeper net model also outperformed other models in predicting the solubility values of a series of newly synthesized compounds for anticancer drug discovery.
As a reliable partner to the world's leading pharmaceutical companies and research institutions, Creative Biolabs brings together the market-leading in silico expertise to build our AI-assisted aqueous solubility prediction service team that can provide the custom professional aqueous solubility prediction.
References