[AI-Powered Automated Protein Structure Preprocessing Platform for Regulatory-Compliant Drug Discovery, 2025 Annual Meeting and International Conference of the KSPST]

AI-Powered Automated Protein Structure Preprocessing Platform for Regulatory_2025 Annual Meeting and International Conference of the KSPST_Poster Drug Design & Development pDDD1.pdf

Download PDF • 1.22MB

Background

Protein structure preprocessing is the foundation of structure-based drug discovery, yet current workflows are highly fragmented. Researchers typically rely on multiple tools such as PyMOL, ChimeraX, and Schrödinger Suite, which leads to complexity, inconsistency, and limited reproducibility. These toolchains often demand advanced expertise in structural biology and computational chemistry, while offering no built-in support for global regulatory requirements (FDA, EMA, PMDA, NMPA, K-FDA). As a result, data integrity is weakened and there is a persistent risk that structure datasets used in docking or virtual screening are not aligned with regulatory expectations.

Objective

To develop an integrated web-AI platform, DockMaster-SONGDO, that automates, standardizes, and validates protein structure preprocessing against international regulatory standards, generating consistent, compliance-ready structure datasets for downstream drug discovery.

Methods

DockMaster-SONGDO is implemented as an in-browser web platform using JavaScript and the Mol* viewer, requiring zero local installation. It provides a six-step preprocessing pipeline with independently executable steps, applicable to targets ranging from small proteins to large complexes (e.g., oncology, neuroscience, immunotherapy, antiviral targets).

Key components include:

1. Global Dataset Presets

   • Fifteen curated biopharma structure datasets pre-validated for quality and regulatory readiness, enabling instant loading of trusted starting structures. AI-Powered Automated Protein St…

2. Automated Preprocessing Workflow

   • Water and ligand removal with real-time monitoring.

   • Metal/ion removal and charge assignment with explicit error tracking.

   • File-format cleanup and final validation/QC with full transparency and auditability. AI-Powered Automated Protein St…

3. Regulatory Compliance Integration

   • Automatic evaluation of resolution, R-factor, and completeness against FDA, EMA, PMDA, NMPA, and K-FDA standards.

   • AI preprocessing engine that applies these standards in real time, producing regulatory-compliant structure datasets.

4. AI-Powered Recommendation Layer

   • State-of-the-art QC metrics (clashscore, Ramachandran statistics, pocket druggability, ligand/metal/water checks, etc.) feed into an optimization and compliance-validation engine.

   • Integration with GPT-5 and Claude 4.5 Sonnet APIs enables AI-driven recommendations for further dataset optimization and structure-based design decisions.

     
     

   Results

   System performance and compliance evaluation show:

   • Processing time: < 5 s for small structures; < 30 s for large complexes.

   • Data integrity: 100% integrity preserved across preprocessing steps in benchmark tests.

   • Compliance: Generated datasets consistently meet resolution, R-factor, and completeness criteria aligned with FDA, EMA, PMDA, NMPA, and K-FDA guidelines.

   • Workflow quality: Transparent, auditable logs provide full traceability from raw PDB input to regulatory-ready output.

     
     

   Conclusion

   DockMaster-SONGDO transforms protein structure preprocessing from a fragmented, expert-dependent, and non-standardized process into an automated, transparent, and regulation-aware pipeline. By unifying data quality control, regulatory compliance checks, and AI-driven recommendations in a single web-AI platform, it delivers reproducible, high-integrity, and submission-ready structure datasets that can be directly leveraged in structure-based drug discovery programs.