bioq: a unified, agent-native command-line interface to a fleet of AI drug-discovery methods
Abstract
Artificial intelligence methods now span the drug-discovery pipeline—structure prediction, de novo design, docking, affinity estimation, and ADMET—yet each tool ships with its own, often incompatible, GPU software stack, so chaining several into a workflow requires reproducing conflicting software environments and access to datacenter-class hardware that most labs do not have. bioq is a dependency-light command-line client backed by bioq-services (a control-plane gateway plus a growing fleet of 38+ containerized drug-discovery tools spanning 7 discovery stages and 6 molecular modalities). The bioq CLI is self-describing and consistent across the fleet of computational tools, giving researchers and coding agents uniform access to any tool from a laptop, with no local model code, CUDA setup, or cloud configuration, running on serverless GPUs billed per job. The interface–gateway–services architecture makes bioq an execution substrate for automated, agent-driven discovery. bioq and bioq-services are open source under the MIT License and available at https://github.com/wolfsonliu/bioq and https://github.com/wolfsonliu/bioq-services. bioq runs on Python ≥ 3.10 with httpx as its only runtime dependency; services run as Linux containers and are self-hostable via the provided local deployment scripts (alongside scripts for Alibaba Cloud Function Compute). Install, quickstart, test data, and self-hosting instructions are in the repository. The project is actively maintained and will continue to receive updates to both features and services.
