Realizing the promise of machine learning in precision oncology: expert perspectives on opportunities and challenges

Background: The ability of machine learning (ML) to process and learn from large quantities of heterogeneous patient data is gaining attention in the precision oncology community. Some remarkable developments have taken place in the domain of image classification tasks in areas such as digital pathology and diagnostic radiology. The application of ML approaches to the analysis of DNA data, including tumor-derived genomic profiles, microRNAs, and cancer epigenetic signatures, while relatively more recent, has demonstrated some utility in identifying driver variants and molecular signatures with possible prognostic and therapeutic applications. Methods: We conducted semi-structured interviews with academic and clinical experts to capture the status quo, challenges, opportunities, ethical implications, and future directions. Results: Our participants agreed that machine learning in precision oncology is in infant stages, with clinical integration still rare. Overall, participants equated ongoing developments with better clinical workflows and improved treatment decisions for more cancer patients. They underscored the ability of machine learning to tackle the dynamic nature of cancer, break down the complexity of molecular data, and support decision-making. Our participants emphasized obstacles related to molecular data access, clinical utility, and guidelines. The availability of reliable and well-curated data to train and validate machine learning algorithms and integrate multiple data sources were described as constraints yet necessary for future clinical implementation. Frequently mentioned ethical challenges included privacy risks, equity, explainability, trust, and incidental findings, with privacy being the most polarizing. While participants recognized the issue of hype surrounding machine learning in precision oncology, they agreed that, in an assistive role, it represents the future of precision oncology. Conclusions: Given the unique nature of medical AI, our findings highlight the field's potential and remaining challenges. ML will continue to advance cancer research and provide opportunities for patient-centric, personalized, and efficient precision oncology. Yet, the field must move beyond hype and toward concrete efforts to overcome key obstacles, such as ensuring access to molecular data, establishing clinical utility, developing guidelines and regulations, and meaningfully addressing ethical challenges.