OralTwin: In-Silico Digital Twin of the Human Oral Microbiome

Overview:OralTwin is an in-silico digital twin framework for the human oral microbiome. This repository establishes the core computational engine, coupling microbial genomics, dynamic host biophysics, and molecular chemistry to simulate bacterial population kinetics and therapeutic responses. What Has Been Implemented & Validated in this Release:1. Genomic Substrate: Integrates 261 curated NCBI whole-genome assemblies across key oral species (P. gingivalis, S. mutans, L. salivarius) balanced in an exact 1:1:1 ratio.2. Dynamic Host Microenvironment: Mathematically models physiological oral perturbations, including the Stephan acid-base curve (pH 4.5–8.5), subgingival periodontal pocket anoxia (0–100% oxygen tension), and salivary clearance.3. Molecular Chemical Representations: Encodes dental therapeutics into dense 384-dimensional continuous feature vectors using the pre-trained ChemBERTa-77M transformer.4. AMR Mutagenesis Testing: Simulates genetic mutations and resistance escape mechanisms (e.g., Delta-pfor metronidazole activation loss and pbp2b modifications). Empirical Validation Benchmarks:• Predictive Accuracy: R2 = 0.9813 (RMSE = 0.0576) on unseen test strains.• Independent External Validation: R2 = 0.9815 (RMSE = 0.0516) on 389 quarantined NCBI genomes (15,171 samples).• Clinical Concordance: 100% alignment with published CLSI breakpoints and human gingival crevicular fluid (GCF) therapeutic ranges. Development Roadmap Toward a Fully Integrated Digital Twin:This foundational release serves as the core baseline. Ongoing and planned development will systematically expand the framework toward a complete, fully integrated oral digital twin by:• Expanding taxonomic coverage to encompass secondary bridge colonizers (e.g., Fusobacterium nucleatum) and complex polymicrobial biofilms.• Integrating diverse patient phenotypes, including diabetic microenvironments, immunocompromised hosts, and salivary xerostomia.• Simulating multi-drug regimens, localized controlled-release delivery, and comprehensive host-microbe metabolic interactions. Repository Package Contents:• best_oral_digital_twin.pt: Trained PyTorch neural network model weights.• train_digital_twin.py: Complete training, inference, and simulation pipeline.• Oral_Microbiome_Digital_Twin_Scientific_Manuscript.pdf: Comprehensive technical report and benchmark documentation.