Webserver Integrated Machine Learning Platform for Predicting Polymer Imprinting Factors in Cancer Therapeutics

Abstract Molecularly imprinted polymers (MIPs) is a class of synthetic receptors capable of specifically recognizing and binding target molecules by forming a molecularly defined binding site. These polymers have been known for its application in drug delivery, clinical diagnosis and biosensing. However, the design and optimization of MIP systems often require extensive experimental trial and error. In this work, we attempted to develop a webserver for the computational design, optimization, visualization of anticancer drug-based MIP systems. The platform enables the user to provide the molecular structures of the template anticancer drug, functional monomers, and crosslinkers in SMILES format, parameters such as the concentrations of template, monomers, crosslinker and monomer molecular weight. Users can customize the number of monomers and crosslinkers to develop virtually different polymer compositions and their predicted imprinting factor can be calculated. The webserver provides real-time three-dimensional visualization of the MIP polymer network interactions. Furthermore, we can predict the properties of bond-length distributions and binding-capacity estimations to analyze molecular interactions and structural characteristics of the designed systems. The imprinting factor (IF) of a molecularly imprinted polymer is predicted using a machine-learning based model as a parameter to evaluate the relative binding performance and selectivity of molecularly imprinted systems. Scientific Contribution: Based on our previous machine-learning algorithm based study for predicting imprinting factors in molecularly imprinted polymers [https://doi.org/10.47852/bonviewJCCE62027832], this work integrates molecular structure based input, virtual polymer composition, three-dimensional structural visualization, binding-capacity assessment, and machine-learning-based prediction of the Imprinting Factor (IF) into a single web-based platform specifically designed for anticancer drug-based MIP systems. It enables the preliminary analysis and optimisation of the MIP formulations before experimental synthesis. We can integrate computational prediction to bridge the gap between conventional strategies and transform experimental optimization. By integrating machine learning based models relative binding efficiency and selectivity can be predicted. The developed webserver differs from conventional MIP optimization workflows. It provides data-driven support to rational optimization of preliminary design of anticancer drugs based MIPs.