proc.bio
Version-controlled bioprocesses with agentic review and a marketplace
Product and process design in biomanufacturing spans scales from molecules to supply chains, generating digital assets across many teams, tools, and formats. As scientific complexity and revision history accumulate, this asset pile becomes too large and fragmented for frequent holistic review. Costly design gaps persist until they surface as late-stage scale-up failures that cost months and millions. When Amyris scaled farnesene production, evolutionary pressure favoring non-producers emerged from system-level dynamics that strain engineering alone could not anticipate. Catching it would have required a cross-scale view of the design, continuously updated as the project evolved.
This work presents proc.bio, version-control-based infrastructure that organizes heterogeneous bioprocess assets and their provenance into a compact, cross-scale representation for agentic review. Assets spanning the full biomanufacturing hierarchy can total 10⁶ to 10⁹⁺ token-equivalent context; a semantic ETL pipeline distills versioned assets into cross-scale data products: compact 10³–10⁴ token views that surface relevant context, assumptions, and evidence for human review, model training, and agent inference. Every pull request triggers a coordinated review. A reasoning model orchestrates specialist agents across three tiers: molecular and reaction feasibility (GNNs, thermodynamic plausibility), process performance (kinetics, FBA, scale-up reconciliation), and techno-economic and process hazard analysis. The fleet returns risk-ranked annotations before work reaches the bench. Merge gates are configurable: hard blocks for regulated contexts, soft annotations for exploratory work.
Accumulated commit history is itself an asset: the provenance record encodes design decisions that agents can mine across scales. Conflicting assumptions and parameter drifts invisible in any single snapshot become visible in aggregate. Validated models, datasets, and manufacturing capacity are published, discovered, and transacted through an integrated marketplace with enforced provenance, license-fee processing, and contract templates that lower barriers to data exchange.
This poster, presented at SynBioBeta '26, reflects the proposed architecture and key user flows, from asset ingestion and cross-scale context distillation to agentic PR review and marketplace integration.