Protein Leverage Satiety Model

ISEF Category: Biomedical and Health Sciences

Subcategory: Nutrition and Natural Products  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

A meal can feel filling and still leave your body asking for more protein. That mismatch can change how much you eat next. If you can measure that pattern in public satiety trials, you can turn a nutrition claim into a prediction model. Then you can test the model on your own meals with a short, controlled intake study.

What Is It?

Satiety means how full you feel after eating and how long that fullness lasts. Protein leverage is the idea that people may keep eating when a meal has too little protein, even if the calories are already high. You can think of protein like the anchor in a boat. When the anchor is light, the meal may not hold hunger in place for long.

In this project, you pull public satiety trials from ClinicalTrials.gov and PubMed, extract the same fields from each one, and build a model that predicts fullness or later intake from protein-related features. Then you test that model in a 14-day self-study, where you keep your own food pattern as controlled as you can and compare the model's prediction with your actual hunger and intake logs.

Why This Is a Good Topic

This topic works well because you can study real human data without a lab bench. The question connects to meal planning, body weight, and sports nutrition, but you can still measure it with public trials, food logs, and simple stats. You also get a clear finish line, a model that either predicts satiety better than a calorie-only version or does not.

Research Questions

• How does protein density in a meal affect next-meal hunger ratings across public satiety trials?
• What is the effect of protein percent of calories on total daily energy intake after adjusting for fiber and energy density?
• Does a model with protein, fiber, and energy density predict satiety better than a model with calories alone?
• To what extent do liquid and solid protein sources change satiety predictions in public trials?
• Which study features, such as age group, body mass index, or meal timing, weaken or strengthen the protein-satiety link?
• How does your model's prediction compare with your own 14-day intake and hunger logs?

Basic Materials

• Laptop with internet access.
• Spreadsheet software such as Google Sheets or Excel.
• Free reference manager such as Zotero.
• Digital kitchen scale with one gram resolution.
• Measuring cups and spoons.
• Food diary app or paper log.
• Body weight scale.

Advanced Materials

• Institutional access to full-text nutrition journals.
• R with the metafor, tidymodels, and ggplot2 packages.
• Python with pandas, statsmodels, and scikit-learn.
• REDCap or Qualtrics project access for repeated hunger surveys.
• Nutrition Data System for Research or another institutional diet-analysis platform.
• High-precision food scale with 0.1 g resolution.
• Fixed-lighting food photo setup.

Software & Tools

• R: Runs the meta-analysis, regression models, and sensitivity checks.
• Python: Cleans extracted trial text and helps automate data preparation.
• Zotero: Stores papers, extracts citations, and keeps your source list organized.
• Google Sheets: Tracks extracted variables, meal logs, and summary metrics.
• JASP: Gives you a point-and-click backup for quick statistical checks.

Experiment Steps

1. Define one satiety outcome and one protein metric so every study gets scored the same way.
2. Build a coding sheet for each trial, including sample type, protein dose, hunger measure, and key controls.
3. Choose a baseline model and a fuller model so you can compare protein-only prediction against an adjusted version.
4. Plan your validation rule, such as leaving one study out or splitting trials into train and test groups.
5. Design your 14-day self-study around repeatable meals, fixed logging times, and the same outcome measure you use in the meta-analysis.

Common Pitfalls

• Mixing hunger scores from different scales, which makes the pooled data look cleaner than it is.
• Using protein grams in one study and protein percent of calories in another without converting them to the same unit.
• Forgetting fiber, energy density, or meal timing, which can make protein look like the only driver.
• Fitting too many predictors to too few trials, which gives a model that sounds strong but fails on new data.
• Letting your own 14-day protocol drift, such as changing sleep, exercise, or meal timing, which blurs the protein signal.

What Makes This Competitive

A stronger version does not stop at a simple regression. You compare study types, control for baseline protein, fiber, and energy density, and test whether your model still works when you leave out one trial at a time. A careful single-subject phase adds a second layer, because you can compare the model's prediction against your own intake data under a planned diet pattern.

Project Variations

• Test whether protein density predicts satiety better in breakfast trials than in lunch or dinner trials.
• Compare animal-based and plant-based protein sources to see whether source type changes the satiety signal.
• Build a simpler model that uses only protein and fiber, then compare its accuracy with a fuller model that adds energy density and meal form.

Learn More

• PubMed: Search review articles and trial reports on satiety, protein leverage, and appetite regulation.
• ClinicalTrials.gov: Find registered human trials that measure fullness, hunger, or ad libitum intake.
• USDA FoodData Central: Check protein, energy, and fiber values for the foods in your protocol.
• NCBI Bookshelf: Read free chapters on nutrition, appetite control, and human metabolism.
• metafor package: Learn fixed and random effects meta-analysis from the package site and vignettes.
• MIT OpenCourseWare: Study introductory statistics and regression through free lecture notes and videos.