Modeling Antibiotic Norms and AMR Spread

ISEF Category: Computational Biology and Bioinformatics

Subcategory: Computational Evolutionary Biology  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A hospital can change its resistance trends without a new drug ever appearing. That is because people change behavior, and bacteria respond to the behavior, not just the medicine. If prescribing habits spread through a clinician network, the microbial gene pool can shift too. Your project asks how fast that chain reaction can happen.

What Is It?

This topic uses an agent-based model, which means you build a simulation made of many individual decision-makers. In this case, the agents are clinicians. Each one can follow a prescribing norm, copy nearby peers, or change behavior after seeing local outcomes. Think of it like a crowd where each person decides whether to wear a jacket. If enough people copy each other, the whole room changes fast.

The biology side comes from antimicrobial resistance, or AMR. AMR means bacteria carry gene variants, called alleles, that help them survive antibiotic pressure. Public hospital genomic surveillance can measure how those alleles change over time. Your model tries to connect the social pattern, how prescribing norms spread, with the biological pattern, how resistance alleles rise or fall. That gives you a bridge between human behavior and microbial evolution.

Why This Is a Good Topic

This is a strong science fair topic because you can test real variables, not just describe a problem. You can change how fast prescribing norms spread, how strongly clinicians copy each other, or how quickly resistance affects future choices. The project connects to a real public health issue, antibiotic resistance, and uses data that hospitals and public databases already track. You can also learn simulation design, parameter testing, and model validation, which are useful research skills.

Research Questions

• How does the rate of clinician-to-clinician norm adoption change the speed of AMR allele-frequency shifts?
• What is the effect of stronger social influence among clinicians on long-term antibiotic prescribing patterns?
• Does adding feedback from local resistance data slow the spread of high-prescribing norms?
• To what extent do different clinic network shapes change the final level of resistance in the simulation?
• Which prescribing rule produces the smallest rise in AMR allele frequency over time?
• How does the initial fraction of cautious prescribers affect the model's resistance trajectory?

Basic Materials

• Laptop or desktop computer with at least 8 GB RAM.
• Python installed with Jupyter Notebook.
• Anaconda or Miniconda for package management.
• Python packages: NumPy, Pandas, Matplotlib, and SciPy.
• Spreadsheet software for data logging and cleanup.
• Public AMR surveillance data from NIH, CDC, or hospital reports.
• Reference papers on antibiotic prescribing behavior and resistance modeling.

Advanced Materials

• Access to a university workstation or cloud compute environment.
• Python with Mesa for agent-based modeling.
• NetworkX for clinician network construction.
• Statsmodels or scikit-learn for regression and sensitivity analysis.
• Git for version control and reproducible workflows.
• Public hospital genomic surveillance datasets or curated allele-frequency tables.
• Optional Bayesian modeling tools such as PyMC for uncertainty analysis.

Software & Tools

• Python: Builds the agent-based model and runs parameter sweeps.
• Jupyter Notebook: Lets you document code, figures, and notes in one place.
• Mesa: Helps you create and manage agent-based simulations in Python.
• NetworkX: Builds clinician contact networks and tests how network shape changes diffusion.
• ImageJ: Measures chart or figure values if you need to digitize published plots.

Experiment Steps

1. Define the clinician decisions you want each agent to make, then choose one behavior rule to test first.
2. Build a simple network structure, such as random, clustered, or scale-free, to represent how prescribing norms move between clinicians.
3. Link prescribing pressure in the model to a resistance outcome so you can track both behavior and allele frequency over time.
4. Add a validation plan that compares your simulated trend against public surveillance summaries or published resistance curves.
5. Plan a sensitivity analysis so you can see which parameters change the result the most.
6. Decide how you will compare model versions, then pick statistical tests that can separate real effects from random noise.

Common Pitfalls

• Treating every clinician as identical, which hides the effect of network position on norm spread.
• Using resistance data without matching the time scale of the simulation, which makes the allele trend look unrealistic.
• Changing too many parameters at once, which makes it impossible to tell what caused the result.
• Confusing prescribing frequency with resistance frequency, which can blur the link between behavior and biology.
• Skipping validation against public data, which leaves you with a model that runs but does not reflect real patterns.

What Makes This Competitive

A strong version of this project does more than run a simulation. It tests multiple network structures, compares them with real surveillance patterns, and checks whether one social rule predicts resistance better than another. You can also add uncertainty analysis, so your results show which conclusions stay stable across repeated runs. That kind of careful modeling makes your work look like real research, not just coding practice.

Project Variations

• Model pediatric, surgical, or ICU prescribing norms separately, then compare which ward spreads resistance fastest.
• Swap the clinician network for a hospital-level network and test how transfers between units change AMR patterns.
• Use published prescription audit data instead of synthetic behavior rules to see whether real-world inputs improve the model fit.

Learn More

• NIH PubMed: Search for review articles on antibiotic stewardship, behavioral diffusion, and antimicrobial resistance modeling.
• CDC Antibiotic Resistance Laboratory Network: Find public summaries of resistance surveillance and hospital trend reports.
• NCBI Pathogen Detection: Explore genomic surveillance resources and read about allele tracking in bacterial populations.
• MIT OpenCourseWare, Introduction to Computer Science and Programming Using Python: Review Python basics and simulation logic through free lecture materials.
• Molecular Biology of the Cell: Use the library or a school copy for background on evolution, selection, and mutation concepts.