BacSafe: A comprehensive database for integrated assessment of bacterial biosafety

Bacterial pathogens account for an estimated 7.7 million deaths annually and rank as the second leading cause of mortality worldwide, while bacterial contamination in cleanrooms of healthcare and industrial facilities—though less frequent—can trigger serious public health crises, with over 40% of food recalls and 31% of pharmaceutical recalls in the U.S. attributed to bacterial contamination. Biosafety risk, however, cannot be captured by any single metric: clinical risk (driven by virulence and antimicrobial resistance) and environmental risk (driven by stress tolerance, biofilm formation, and persistence) are not inherently correlated, as illustrated by the contrast between the fastidious but highly virulent Neisseria meningitidis and the rarely pathogenic yet environmentally hardy Bacillus subtilis. Existing resources capture only theoretical risk from genomic or phenotypic predictions, while real-world evidence such as recall events and outbreak investigations remains scattered across disparate platforms. To bridge this gap, the authors developed BacSafe (http://bacsafe.dmicrobe.cn/), a knowledge database that uniquely characterizes both clinical and environmental biosafety risk for bacterial species.

Data Content and Risk Framework. BacSafe covers 3,902 genera and 20,630 species, with each entry encompassing 89 biosafety-related attributes organized into two complementary modules. The environmental risk module combines theoretical traits (morphology, physiology, biochemistry, disinfectant tolerance, water activity requirements, and biofilm-forming capacity) with empirical indicators, including all documented FDA recall events in food and pharmaceutical production and detection frequencies within cleanrooms—derived from published literature on UK facilities and an in-house dataset from 313 Chinese pharmaceutical plants collected over four years, plus contamination data from surgical suites and space stations. The clinical risk module integrates genome-based annotation of virulence and AMR genes (with prevalence metrics across all sequenced genomes of each species) with empirical indicators such as biosafety levels assigned by national agencies, pathogenicity ratings, host and disease records, and a PubMed-based rating quantifying virulence, resistance, and transmissibility through literature counting. In total, the database is compiled from 711,344 genome assemblies, 2,517,153 PubMed publications, and 20,437 recall records.

BacSafe offers multiple access pathways: standard search by species or genus name, attribute-based search for complex multi-criteria queries, and an integrated AI assistant built by fine-tuning the DeepSeek-R1 large language model within a retrieval-augmented generation (RAG) architecture that combines Text2SQL structured queries with semantic retrieval. On a 100-question benchmark independently evaluated by a pharmaceutical quality-control specialist, a clinical laboratory technician, and an infectious disease specialist (Fleiss' Kappa = 0.861), the BacSafe assistant achieved 81% accuracy, outperforming ChatGPT-5 (69%), Gemini-2.5 Pro (66%), and DeepSeek with or without web search (43–44%). Beyond querying, BacSafe provides decision-tree-based tools that operationalize the evaluation of "objectionable microorganisms" in non-sterile products and cleanroom environments, implementing algorithms from the Chinese Pharmacopoeia (2025 edition) and PDA Technical Report 67; users input the contaminating species alongside product characteristics, administration route, and patient population to obtain context-specific assessments. Additional computational tools accept user-uploaded whole-genome sequences, returning taxonomic identification, predicted environmental adaptability traits, and annotations of virulence factors and AMR genes.

By consolidating dispersed aspects of biosafety knowledge into a structured, multidimensional, and context-dependent resource, BacSafe moves beyond the single unified risk scores of existing databases and enables tailored risk assessment for diverse users—pharmaceutical quality-control specialists prioritizing environmental persistence and disinfectant tolerance, and clinical microbiologists focusing on virulence and AMR profiles. The database distinguishes itself from resources such as BacDive, MetaTraits, gcPathogen, and NCBI Pathogen Detection by uniquely integrating bibliometric analysis, regulatory reports, cleanroom isolation data, and industry recall events with AI-assisted querying and industrial risk-assessment tools. BacSafe is open-access under CC BY-NC 4.0, updated biannually, and serves as a vital resource for bacterial risk assessment in clinical, industrial, and research settings. Future development will expand taxonomic coverage to viruses, fungi, and protozoa, and integrate autonomous AI agents capable of intelligent biosafety risk evaluation coupled with targeted control recommendations—establishing a new paradigm of intelligent, proactive microbial risk management