زينب سودايي

Solubility is one of the fundamental challenges in drug design and development that affects topics such as biological activity, clinical trials, toxicity testing, pharmacokinetics, and formulation. Oral medications are considered as the main option in the development of drugs due to their ease of use, but however the basic problems of oral dosage are related to their poor bioavailability, the most common cause of which is low solubility. In the field of drug research and development, the recognition and eva‎luation of the solubility of new molecules can reduce the workload by removing compounds with undesirable properties. Experimental methods and simulation of molecular dynamics are among the solutions used to determine solubility, but the methods are time-consuming and expensive. Machine learning is a subset of artificial intelligence that has the ability to infer and find implicit relationships from data in a statistically accurate way without knowing the problem. The study used a variety of machine learning algorithms including linear regression methods, decision tree, random forest, and group methods to predict the solubility of pharmaceutical compounds of acidic, basic and neutral nature from various structural classifications such as pteridine, theophylline, barbiturate, etc. Along with quantum characteristics, molecular and structural dynamics extracted from calculations and papers. That boosting gradient and random forest methods provided good performance in predicting solubility. In order to improve this project in the future, we are going to use deep neural network models to make more accurate predictions and to examine the impact of thermodynamic quantities on solubility.

يادگيري ماشين , شيمي محاسباتي , حلاليت , ديناميك مولكولي , مكانيك كوانتومي

machine learning , computational chemistry , solubility , molecular dynamics , quantum mechanics

zeinab sodaei

seyed majid hashemianzadeh