Deep learning-based design of cell-targeting ligands enables targeted and functional siRNA delivery to lung cell subsets

<bold>Background:</bold> Oligotherapeutics are a promising tool for treating diseases that cannot be modulated by standard therapeutics. Conjugation of stabilized siRNAs to a ligand capable of mediating receptor-mediated endocytosis is a clinically viable strategy for delivery to hepatocytes. Receptors capable of mediating uptake of siRNA conjugates in lung tissue, however, have limited clinical validation and identification of these receptors remains challenging. <bold>Approach:</bold> By combining single-cell transcriptomics with an in silico model of receptor internalization kinetics, we have identified receptors that are expressed on diseased lung cells and are rapidly internalized upon binding. Then, using a series of physics-aware deep learning models (constructed via datasets of millions of receptor-ligand interactions), we design protein-based ligands that can be conjugated to a siRNA payload. We have designed high affinity ligands against multiple receptors in lung tissue, enabling the functional and targeted delivery of siRNA to different cell types. We demonstrate functional and selective delivery to lung epithelial tissue & resident immune cells from systemically administered compounds, and have shown this approach can knockdown clinically relevant genes in those cells. <bold>Conclusion:</bold> The rapid discovery and interrogation of novel ligand-receptor pairs using deep learning methods offers a promising approach for achieving targeted functional delivery in previously inaccessible tissue & cell types. These methods, when combined with prior achievements in siRNA chemistry, provide a toolkit for the rapid development of oligo conjugate medicines for patients with unmet needs.