Highlights A novel comprehensive cloud-native MLOps platform covers the entire preclinical drug discovery process end-to-end. Flexible integration strategies bridge a fragmented digital ecosystem and deliver AI value to scientists for informed decisions. Decoupling of machine learning and operational code enhances reliability and R&D specialization. Abstract The increasing application of machine learning (ML) and artificial intelligence (AI) in drug discovery necessitates robust and scalable solutions for operationalizing these technologies. This paper presents the development and implementation of a cloud-native Machine Learning Operations (MLOps) platform designed to accelerate the delivery of production-ready, scalable, and reliable AI models within the pharmaceutical drug discovery environment. The platform addresses challenges associated with operationalizing AI in complex research and development settings, emphasizing the critical interplay between well-defined business requirements and robust technical capabilities. Foundational business needs, such as ensuring model trustworthiness, data traceability, and rapid experimental iteration, directly influenced the technical architecture, including choices around orchestration (Kubeflow Pipelines), real-time serving (KServe), data integration, and monitoring. Our comprehensive MLOps approach demonstrates how strategic alignment of business objectives with scalable technical solutions can unlock significant value, showcasing practical applications across diverse ML workflow types. The platform fosters operational excellence, accelerates research in a cost-effective manner, and provides valuable insights and a proven methodology for bridging the gap between scientific innovation and large-scale AI deployment in various industries.
Koletou et al. (Thu,) studied this question.