Deep Learning-Driven Co-Assembly of Naturally Sourced Compound Nanoparticles for Potentiated Cancer Immunotherapy

Abstract

Co-assembly of excipient-free nanoparticles has emerged as a promising drug delivery platform due to their high drug-loading capacity, ease of preparation, and ability to achieve combination therapeutic effects. However, the absence of systematic design strategies has hindered their broader application. In this study, a deep learning platform, Gramord, is developed to rationally design the excipient-free anti-tumor nanoparticles of nature-sourced compounds. A comprehensive database of excipient-free nanoparticles is first built and used to train Gramord for predicting self-assembly compatibility. By screening 1800 naturally-derived small molecules and their derivatives, the compound pairs capable of forming excipient-free nanoparticles are identified. Leveraging the advantage of oridonin (Ori) for inducing apoptosis of tumor cells and cepharanthine (Cep) for eliciting immunogenic cell death of tumor cells, the Ori-Cep pair for preparing the self-assemble nanoparticles (namely OCN) is subsequently selected. Using a mouse model of CT26 colorectal tumor, it is demonstrated that the systemically administrated OCN specifically accumulate at the tumor sites, and regress tumor growth by inducing anti-tumor immunogenicity and recruiting tumor-infiltrating cytotoxic T lymphocytes. This study highlights the application of artificial intelligence in designing excipient-free nanomedicine, offering a scalable and cost-effective approach to expanded therapeutic options.