AI-Driven Prediction of Antimicrobial Resistance via Universal Antimicrobial Mechanism Model (Pfizer)

1. Research Background and Summary
Antimicrobial Resistance (AMR) is a top-tier global health threat, particularly concerning in Gram-negative bacteria such as those causing Carbapenem-Resistant Enterobacteriaceae (CRE) infections. I have observed that current clinical practices are fundamentally reactive: traditional testing is slow and can only detect resistance that is already established, offering no foresight into emerging threats.
My research directly addresses this critical gap. I am developing an AI-Driven Antimicrobial Resistance Early Warning System, which I call the Universal Antimicrobial Mechanism Model (UAMM).
My methodology integrates a three-part "Predict-Explain-Discover" (P-E-D) framework:
1. Predict: I will use high-end deep learning models (Transformer architecture) to predict the exact level of drug resistance (MIC values) based purely on the genetic sequence of a mutation.
2. Explain: I will employ structural biology tools (AlphaFold) to interpret why the mutation is effective.
3. Discover: I will integrate these AI-predicted parameters into Virtual Cell (VCell) simulations to dynamically model the complex fight between the antibiotic and the bacterium inside the cell.
My initial strategic focus is on Klebsiella pneumoniae. I will utilize the high-quality, sample-level data from Vivli to ensure my prediction model is robust and biologically valid before expanding the system to address more complex superbugs.
2. The Public Health Impact
How our proposed research will: Help improve patient outcomes
Our system provides the foundation for rapid, precision-guided therapy for patients with severe infections. We can assess the resistance risk from genetic data in hours—significantly faster than the days required for traditional lab tests. This rapid insight allows clinicians to select the correct, effective antibiotic immediately upon patient admission. By preventing delays in appropriate treatment, we aim to directly reduce patient morbidity and mortality rates associated with severe AMR infections.
How our proposed research will: Strengthen stewardship
We strengthen antibiotic stewardship by providing data-driven certainty. The high accuracy and explainability of our AI models will empower hospital infection control teams to create more informed empirical treatment protocols. By accurately predicting which drugs will fail against specific strains, we reduce the unnecessary use of broad-spectrum antibiotics, thereby preserving the efficacy of existing drugs for longer and actively combating the evolution of new resistance mechanisms.
How our proposed research will: Inform public health practice
Our research provides a proactive, automated risk-stratification tool for public health surveillance. The UAMM system is designed to assess any newly identified resistance mutation (e.g., in databases or from WGS surveillance data). If the model predicts that a novel variant poses a high threat, public health agencies can receive an early, actionable warning. This allows them to proactively implement containment strategies, strengthen border monitoring, and allocate resources efficiently before a high-risk strain causes an outbreak.
How our proposed research will: Strengthen health systems
The final outcome of this project is the delivery of a low-cost, high-efficiency digital decision-support tool. I am building this system using full-stack engineering principles (Docker containers and a web platform). This ensures that the technology can be easily integrated into existing health system infrastructure globally, particularly in resource-limited settings. By automating complex genomic analysis and providing clear resistance predictions, we increase the diagnostic capacity and overall resilience of health systems to face the persistent and unpredictable challenge of AMR.