Computational design and development of small protein inhibitors

Abstract With the advent of small, highly stable miniproteins and the ability to generate easily more, methods and prototypes for their applications both computationally as well as experimentally are needed. As part 1, we will investigate the potential of miniprotein-based fusion inhibitors. We will target multiple class I fusion proteins and evaluate their potential as treatment using two animal models. Many pathogenic viruses, including influenza, Ebola, coronaviruses and the Pneumoviridae, rely on class I fusion proteins to fuse the viral and cellular membranes. Their fusion proteins change from the metastable pre- to the more energetically favorable post-fusion state which is thought to drive membrane merging. The postfusion structure gives insight into how a potential intermediate state can be inhibited. We propose the development of a platform that takes advantage of this phenomenon and extracts information from the post-fusion state structure to develop fusion inhibitors. We are targeting CoV-2, Respiratory Syncytial Virus (RSV), Nipah virus and Middle East Respiratory Syndrome (MERS). We have generated several candidates against CoV-2 and RSV. We will evaluate their neutralization potential first in cell culture and then analyze the most potent version in animals. We will explore treatment window, dosage, and delivery methods, linkage to co-localizers and their immunogenicity. We have established a proof-of-concept illustrating that we can computationally design inhibitors, verify, and optimize these binding using yeast surface display: preliminary data demonstrates, an inhibitor generated with this pipeline neutralizes the live CoV-2 virus in cell culture. We believe the proposed framework will be applicable for more class I fusion viruses and provide a general protocol for the development or optimization of new biologics based on miniproteins. Furthermore, our platform will provide technology for pandemic preparedness. Second, we are proposing the development of two independent deep learning-based protein design methods to improve stability and to design protein-protein interactions from scratch. For the generation of de novo protein- protein interactions (PPIs), we have previously developed a prototype algorithm. Sequence and interface recovery indicate high accuracies for predicting hotspot interactions, both, for their surface locations, as well as their identities. We previously established a stability predictor based on the evaluation of 31,000 designed miniproteins. We aim to expand on our success and built a more complex neural network-based algorithm that can guide re-design. We will integrate other published dataset, evolutionary information while also feeding back any information obtained while optimizing for stability under strenuous conditions (e.g. low pH, high temperature, high protease concentrations and in serum) of any of our designed proteins. We will ensure an iterative coupling between computation and experimental evaluation.