article Open access
Papers open inside SciRah from the original publisher or repository; SciRah never hosts them.
The Department of Energy’s Systems Biology Knowledgebase (KBase) serves as an open source research platform for reproducible and collaborative computational biology workflows based on FAIR data principles. As AI transforms scientific computing, KBase is evolving beyond workflow execution to a knowledge engine that consumes, produces, and exposes reusable scientific assets. The resulting platform integrates harmonized data resources, computational infrastructure, intelligent research capabilities, and a growing body of reusable expertise captured from community use to create an open and extensible analysis platform for next-generation AI-enabled science. This vision aims to lower barriers to scientific exploration while maintaining transparency, reproducibility, and scientific rigor. This talk highlights these advances through two closely related efforts: development of the KBase Lakehouse and deployment of an AI co-scientist (BERIL) as an initial realization of KBase’s intelligent research capabilities. One core aspect of this vision, the KBase Lakehouse, is a scientific data infrastructure designed to transform fragmented biological datasets into a connected AI-ready data ecosystem. Multi-tenant by design, the lakehouse federates data from KBase, JGI, NMDC, EMSL, NIH, USGS, and other communities. As an initial demonstrator, KBase has deployed BERIL, an AI co-scientist system that operates on the KBase Lakehouse infrastructure. The KBase ecosystem combines the Lakehouse’s diverse harmonized data resources with computational infrastructure, provenance, and community engagement. This enables BERIL to introduce agentic capabilities that together accelerate the cycle of hypothesis generation, analysis, discovery, and communication. Rather than a standalone chatbot, BERIL layers a suite of skills interfacing into the KBase Lakehouse infrastructure that supports transparent, reproducible AI-assisted science through traceable reasoning, reusable scripts, and shared lessons from prior user workflows. In a recent one-day user experience workshop, first-time users of the new platform produced multiple review-ready publications demonstrating how an AI-assisted research ecosystem can dramatically lower barriers to complex scientific discoveries. We conclude by reflecting on the evolving relationship between scientists and AI systems, the infrastructure design patterns required to build sustainable AI systems, and on the broader challenges of building open, trustworthy, and collaborative AI infrastructure for scientific discovery.