Drug-repurposing screen on patient-derived lung cancer organoids with deep learning model

"Background:Lung cancer remains the top cancer-related mortality in most parts of the world, largely due to late-stage diagnosis, metastasis, and resistance to conventional therapies. Recent advances in understanding the role and mechanisms of key driver mutations in lung cancer have facilitated the development of targeted therapies. However, the mechanisms of progression and development of drug resistance in lung cancer treatment is not fully understood, necessitating exploration of novel molecular targets and resistance mechanisms.Methods:This study leverages next-generation sequencing (NGS) and deep learning model to enhance personalized lung cancer treatment.Results:We established patient-derived lung cancer organoids (LCOs) from resected tumors and pleural effusions of 14 patients with non-small cell lung cancer (NSCLC). Histological and genomic analyses revealed prevalent TP53, TTN, MUC16, and FLG mutations. Notably, early-stage tumors exhibited higher tumor mutation burdens (TMB) compared to advanced-stage tumors. Utilizing the DrugCell deep learning model, the efficacy of various anti-cancer drugs on patient-derived LCOs were predicted, followed by cytotoxicity assays to validate the corresponding predictions. The model demonstrated moderate sensitivity (68.28%) and accuracy (59.34%), reflecting its potential applicability in pre-selection of effective treatments. Discrepancies between predicted and actual drug responses underscore the limitations of the model for refinement and incorporation into more advanced in vitro experimental conditions.Conclusions:This study demonstrated the utility of integrating genomic profiling with computational models to predict therapeutic responses, advocating for a shift towards more personalized and cost-effective treatment strategies in lung cancer."