Machine Learning Assisted Prediction of Physical Properties of Cyclotides

Cyclotides are organic molecules that typically contain 28-37 amino acids. They may be isolated from certain plants and have a wide range of biological activity such as being insecticidal, anti-tumor, anti-microbial. Their biological activity and remarkable chemical stability provide an exciting range of potential therapeutic applications. Traditional Molecular Dynamics (MD) studies of cyclotides experience computational limitations due to the large number of atoms in cyclotides. Machine-learning models serve as cost-effective and time-saving computational tools in predicting their physical properties which may be compared to MD calculations.

In this study, we use a set of datafiles created by MD study of Kalata-B1 molecule (a type of cyclotide) to create a feature vector(s) consisting of numerical metrics that capture the chemical interactions of Kalata-B1 molecule. As a part of this study, we plan to generate and process the MD data to engineer predictor variables, to generate test/training data sets, and to develop and test machine learning model(s) that predict the simulational results of MD calculations.