AI‑Driven Multi‑Scale Framework for Probiotic Behavior Prediction

This archive contains all computational resources associated with the manuscript “AI‑Driven Multi‑Scale Framework for Probiotic Behavior Prediction.” It includes curated genomic data, processed protein embeddings, metabolic models, machine‑learning features, trained models, and supplementary validation materials. A detailed README file provides step‑by‑step instructions for reproducing the full computational workflow. It includes: 1. Source Code (GitHub mirror) Scripts for genomic curation, module discovery, PLM inference, metabolic modeling, and supervised learning Implementations of GNN, VAE, transformer‑based attention, and metabolic simulation workflows Environment files and Docker configuration for reproducibility 2. Processed Datasets Curated genomic annotations for 156 probiotic strains Gene‑module matrices (247 modules) Protein embeddings (ESM‑2 and ProtT5) Matrix‑specific metabolic flux outputs (10,248 simulations) Integrated AI‑ready dataset used for machine‑learning models 3. Workflow Documentation Step‑by‑step instructions for reproducing the computational pipeline File descriptions and directory structure Links to the GitHub repository for version‑controlled development 4. Supplementary Materials Benchmarking results for PLM classifiers Module definitions and functional domain assignments Additional figures and tables referenced in the manuscript Directory Structure Probiotic_AI_Framework_Dataset_v1.0/ ├── genomic_data/ │ ├── complete_genomes_156.tar.gz # All 156 genome sequences (FASTA) │ ├── gene_annotations.tar.gz # Prokka, RAST, KEGG annotations │ ├── gene_modules_247.xlsx # Complete module catalog │ ├── phylogenetic_analysis.tar.gz # Trees and evolutionary data │ └── quality_metrics.csv # Assembly statistics and QC data ├── protein_data/ │ ├── protein_sequences_8847.fasta # All analyzed protein sequences │ ├── esm2_embeddings.h5 # ESM-2 embeddings (1280-dim) │ ├── prott5_embeddings.h5 # ProtT5 embeddings (1024-dim) │ ├── functional_predictions.csv # PLM-based annotations │ ├── confidence_scores.csv # Prediction confidence metrics │ └── validation_results.xlsx # Experimental validation data ├── metabolic_models/ │ ├── individual_gems/ # 156 strain-specific GEMs (SBML) │ ├── flux_analysis_results.csv # All 42,640 simulation outputs │ ├── matrix_constraints.json # Environmental parameters │ ├── pathway_annotations.csv # Metabolic pathway mappings │ └── experimental_validation.xlsx # Flux validation data ├── machine_learning/ │ ├── feature_matrices/ # Processed ML input features │ │ ├── genomic_features_247.csv │ │ ├── proteomic_features_847.csv │ │ └── metabolic_features_153.csv │ ├── trained_models/ # Serialized ML models │ │ ├── scfa_prediction_model.pkl │ │ ├── il10_induction_model.pkl │ │ ├── survivability_model.pkl │ │ └── adhesion_model.pkl │ ├── cross_validation_results.csv # Performance metrics │ ├── feature_importance.csv # SHAP analysis results │ └── prediction_outputs.csv # Final strain predictions ├── ai_analysis/ │ ├── gnn_embeddings.h5 # Graph neural network outputs │ ├── vae_latent_coordinates.csv # Variational autoencoder results │ ├── transformer_attention.h5 # Attention weight matrices │ ├── umap_coordinates.csv # UMAP projection coordinates │ └── cluster_assignments.csv # Module cluster memberships ├── validation/ │ ├── experimental_validation.xlsx # Lab confirmation data (45 strains) │ ├── literature_support.csv # Citation evidence for predictions │ ├── benchmark_comparisons.xlsx # Performance vs other tools │ ├── temporal_validation.csv # Independent test results │ └── statistical_analyses.R # Validation statistical tests ├── supplementary/ │ ├── figure_source_data/ # Data for all manuscript figures │ ├── additional_analyses/ # Extended computational results │ ├── method_comparisons/ # Benchmarking data │ └── strain_characterization/ # Individual strain profiles └── metadata/ ├── dataset_description.md # Comprehensive documentation ├── data_dictionary.csv # Variable definitions ├── experimental_protocols.pdf # Validation methods ├── computational_methods.pdf # Algorithm descriptions └── version_history.md # Dataset update log