[Paper Review] Accelerating drug discovery with Artificial: a whole-lab orchestration and scheduling system for self-driving labs

Yao Fehlis, P. Mandel+3 more|ArXiv.org|Apr 1, 2025

Scientific Computing and Data Management3 citations

TL;DR

This paper presents Artificial, a whole-lab orchestration and scheduling platform that unifies lab operations, automates workflows, and integrates AI models (e.g., NVIDIA BioNeMo) to accelerate AI-guided drug discovery in both dry and wet labs.

ABSTRACT

Self-driving labs are transforming drug discovery by enabling automated, AI-guided experimentation, but they face challenges in orchestrating complex workflows, integrating diverse instruments and AI models, and managing data efficiently. Artificial addresses these issues with a comprehensive orchestration and scheduling system that unifies lab operations, automates workflows, and integrates AI-driven decision-making. By incorporating AI/ML models like NVIDIA BioNeMo - which facilitates molecular interaction prediction and biomolecular analysis - Artificial enhances drug discovery and accelerates data-driven research. Through real-time coordination of instruments, robots, and personnel, the platform streamlines experiments, enhances reproducibility, and advances drug discovery.

Motivation & Objective

• Motivate the use of self-driving labs to address high costs, long timelines, and data fragmentation in drug discovery.
• Propose a unified platform that orchestrates workflows, coordinates instruments and personnel, and manages data efficiently.
• Demonstrate integration of AI-driven decision-making (e.g., BioNeMo models) within automated lab workflows.
• Showcase a proof-of-concept case study in dry lab virtual screening to validate AI-model deployment and resource optimization.
• Highlight the end-to-end architecture enabling scalable, secure lab automation across cloud and on-premises environments.

Proposed method

• Describe the Artificial stack with four layers: Web Apps, Services, Lab API, and Adapters/Protocols.
• Explain the orchestration and scheduling engine that plans, executes, and optimizes workflows using heuristics and batching.
• Detail the data management via Data Records and the Digital Twin visualization for real-time monitoring.
• Demonstrate integration of NVIDIA BioNeMo NIM microservices through the LabGateway for secure on-premise or cloud deployment.
• Present a PoC workflow for self-driving virtual screening targeting SARS-CoV-2 main protease, with iterative AI-guided cycles and performance criteria.
• Outline how the platform connects to LIMS/ELN and supports secure, scalable deployment in cloud (EKS/AKS) or local (MicroK8s) environments.

Experimental results

Research questions

• RQ1 How can a whole-lab orchestration platform consolidate lab operations, AI models, and data into a reproducible workflow?
• RQ2 Can AI-driven decision-making be effectively integrated into automated virtual screening and wet-lab workflows to accelerate drug discovery?
• RQ3 What architectural patterns (web apps, services, lab API, adapters) enable scalable, secure, and interoperable self-driving labs?
• RQ4 What are the practical benefits and limitations of deploying AI models like BioNeMo within an orchestrated lab environment?
• RQ5 How does the system perform in a proof-of-concept case study involving dry-lab virtual screening and AI model integration?

Key findings

• Artificial enables automated, AI-guided decision-making and real-time coordination of instruments, robots, and personnel.
• The PoC demonstrates integration of BioNeMo NIM microservices within Artificial to enable self-driving virtual screening.
• In the SARS-CoV-2 main protease use case, three iterative self-driving cycles achieved predefined criteria (binding affinity threshold of -1.4 million and at least ten successful molecules).
• The architecture supports secure data exchange, immutable data records, and seamless data access for training or fine-tuning AI models.
• The platform is designed for deployment across cloud and on-premises environments, leveraging a Digital Twin and lab-wide orchestration to improve reproducibility and throughput.