Lesson 2.1 Macronutrients, Micronutrients, Dietary Patterns

Outcomes (progressive ladder)

• Foundational: Explain the roles of carbohydrates, proteins, and fats; distinguish water-soluble vs fat-soluble vitamins and key minerals; define common dietary patterns.

• Intermediate: Convert foods to macro/micro features (grams, %kcal, density per 1,000 kcal); compute energy needs and macro targets from goals.

• Advanced: Build nutrition-aware feature sets for ML, including quality indices (e.g., fiber density), pattern tags, and culturally diverse cuisine encodings; reason about trade-offs (adherence, budget, sustainability) in recommendation logic.

Concept Primer

Nutrition science studies how foods (matrices of nutrients and bioactives) affect health, performance, and disease risk. AI systems need a crisp representation of food and intake to learn anything useful. That means:

1. Quantities (grams, milliliters) and energy (kilocalories).

2. Composition (macros, micros, fiber, sodium, added sugar).

3. Context (meal timing, culture/cuisine, budget, health status).

Think of a dish as a vector:

Good apps also attach metadata: dish name(s) in multiple languages, cuisine, cooking method, and uncertainty bounds (portion estimates vary!).

Macronutrients (functions, sources, AI-relevant features)

Carbohydrates

• Role: Primary energy; includes starches, sugars, and fiber (non-digestible, key for satiety and glycemic control).

• Sources: Grains (rice, roti), legumes (chana), fruits, starchy veg.

• Feature ideas: total carbs (g), net carbs (carbs − fiber), added sugar (g), glycemic load proxy (carbs_g × portion_factor), fiber density (g per 1,000 kcal).

Protein

• Role: Tissue maintenance, hormones/enzymes; higher satiety per kcal.

• Sources: Dal, chana, tofu/paneer, eggs, fish, lean meats, dairy.

• Feature ideas: protein (g), % kcal from protein, complete vs incomplete sources (tag), leucine estimate (advanced).

Fat

• Role: Energy-dense, essential fatty acids, vitamin transport.

• Sources: Oils/ghee, nuts, seeds, fish.

• Feature ideas: total fat (g), SFA/MUFA/PUFA split if available, omega-3 proxy (fish/nuts tags), SFA %kcal.

Practical guardrails: normalize macros to %kcal for cross-meal comparison; track fiber separately (predicts satiety and adherence).

Micronutrients (high-impact highlights)

Vitamins

• Water-soluble: B-complex, C — limited storage; frequent intake matters.

• Fat-soluble: A, D, E, K — absorbed with fat; risk both for deficiency and excess.

• Feature ideas: per-1,000 kcal densities for vitamin A (RAE), D (IU/µg), C (mg), B12 (µg); fortified flag for cereals/dairy.

Minerals

• Electrolytes: Sodium, potassium — key for blood pressure/fluid balance.

• Bone health: Calcium, magnesium; iron for oxygen transport; zinc for immunity.

• Feature ideas: sodium (mg), Na:K ratio, calcium density (mg/1,000 kcal), iron source type (heme/non-heme tag).

Label uncertainty is common. Many databases miss fatty acid breakdown or vitamin D for regional foods. Carry missingness flags into your model.

Energy Balance & Target Setting (for planners)

1. Basal estimate: (e.g., Mifflin-St Jeor) to get Resting Energy Expenditure (REE).

2. Total Daily Energy Expenditure (TDEE): REE × activity factor.

3. Goal adjuster: deficit/surplus for weight goals or neutrality for maintenance.

4. Macro targets: set as %kcal or g/kg (e.g., protein 1.2–1.6 g/kg/day for active adults; adapt to local guidance and clinical context).

Feature practice: compute %kcal from macro

(similarly carbs ×4, fat ×9)

Dietary Patterns (what they mean to an algorithm)

Patterns encode bundles of choices beyond single nutrients. Represent them as tags + constraints + typical macro ranges. Examples:

• Mediterranean-like: high vegetables/legumes/olive oil, moderate fish, minimal processed meats. Features: high MUFA, fiber density; seafood tag.

• DASH-like: emphasizes fruits/veg/low-fat dairy; limits sodium. Features: Na:K ratio, sodium cap per day.

• Vegetarian/Vegan: excludes meats (and dairy/eggs for vegan). Features: plant protein share, B12 risk flag, omega-3 ALA sources.

• Lower-carb ranges: restricts high-GI starches. Features: net carbs %kcal thresholds, fiber maintained.

• Culturally anchored patterns (e.g., Indian regional): rotis, dals, sabzi, rice varieties, millets. Features: cuisine tag, staple type (wheat/rice/millet), typical oil.

Adherence matters. A slightly less “optimal” plan that matches culture, budget, and schedule will often beat a theoretically perfect one. Encode preference vectors and budget bands as first-class features.

From Food to Features: a repeatable mapping

1. Parse & standardize

   • Map items to a food code (e.g., regional DB + FoodData-style schema).

   • Convert household measures → grams (unit dictionary, density tables).

2. Aggregate macros & micros

   • Sum per item → per meal → per day.

   • Produce both absolute (g, mg) and relative (%kcal, per-1,000 kcal) features.

3. Quality signals

   • Fiber density, Na:K ratio, added sugar %, SFA %kcal, diversity index (unique foods / day or week).

   • Cooking method tags (fried/steamed), ultra-processed flag where available.

4. Context

   • Time (circadian window), eating window (fasting), budget tier, cuisine.

5. Uncertainty

   • Portion confidence; imputed micronutrient flags; barcode vs free-text source.

Pitfalls & How to Handle Them

• Portion guesswork: ask for reference objects (plate/bowl size) or hand size; carry ±CI on grams.

• Mixed dishes: use recipe decomposition (dal makhani → dal + butter/ghee + cream) with regional defaults; allow user overrides.

• Unit chaos: grams vs “servings”; create canonical serving sizes and expose conversions.

• Micronutrient gaps: backfill with recipe rules (e.g., fortified atta), but mark as estimated.

• Added vs intrinsic sugar: many DBs don’t separate—estimate via recipe tags; be transparent.

• One-number traps: Avoid overreliance on calories; track protein and fiber as independent targets for satiety.

AI-Ready Feature Cookbook (copy/paste to specs)

Daily rollups

• kcal_total, protein_g, carb_g, fat_g, fiber_g, sugar_added_g, sodium_mg

• %kcal_protein, %kcal_carb, %kcal_fat

• fiber_per_1000kcal, sodium_per_1000kcal, na_k_ratio

• sfa_pct_kcal, mufa_pct_kcal (if available)

• diet_diversity_7d (unique items / 7d)

Pattern tags

• pattern_mediterranean_like, pattern_dash_like, vegetarian, vegan, lower_carb_range

• cuisine_region (one-hot or embedding), staple_grain_type (rice/wheat/millet), cooking_method_fried

Adherence/context

• meal_timing_entropy, fasting_window_flag, budget_band, prep_time_minutes

Quality flags

• portions_conf_low, micros_imputed, barcode_scan, free_text_parsed

Worked Mini-Examples

Example 1 — Convert a meal to features

“2 rotis (60 g flour total), 1 cup chana masala, 1 tsp ghee, cucumber salad”

• Standardize grams per item using the DB.

• Sum macros; compute %kcal_protein and fiber_per_1000kcal.

• Add tags: cuisine_north_indian, vegetarian, cooking_method_sautéed, added_fat_ghee.

Example 2 — Sodium feature and DASH constraint

• Daily sodium = 2,100 mg; potassium = 3,400 mg → na_k_ratio ≈ 0.62.

• If app targets DASH-like, enforce daily sodium ceiling and prioritize high-K foods (bananas, leafy greens, dal).

Example 3 — Pattern scoring

• Construct a simple Mediterranean score (0–9): points for legumes, nuts, vegetables, whole grains, fish; subtract for processed meats; add MUFA/SFA ratio > 1. Encode as med_score.

How this fuels ML & Planning

• Labels: “hit daily fiber (≥25 g)” or “sodium over target.”

• Supervised features: %kcal_protein, fiber_per_1000kcal, timing features → predict satiety/adherence.

• Unsupervised: cluster weekly macro ratios + diversity to personalize plans.

• RL/bandits: actions = suggest high-fiber swaps; rewards = confirmations + fiber target attainment with caps on nudges.

Practice (self-check)

1. Convert your last meal into grams, kcal, %kcal, and fiber density. What two swaps would raise fiber without raising kcal > 100?

2. Define a label for “met protein goal” that avoids leakage; specify your windows.

3. Draft a pattern score (0–5) for a culturally relevant diet you know; list which features drive it.

4. Create a feature list for sodium-aware planning, including constraints and a Na:K target.

5. List the uncertainty flags you’d store for a plate-photo entry and how the UI should react.

Glossary (targeted)

• %kcal from macro: Macro kcal / total kcal × 100.

• Fiber density: grams of fiber per 1,000 kcal (quality indicator).

• Na:K ratio: Sodium to potassium intake; lower is generally favorable.

• Diet diversity index: Unique foods over a window; proxy for micronutrient breadth.

• Pattern score: Composite indicator reflecting adherence to a dietary pattern.

• Fortification: Nutrients added to foods (e.g., B12 in plant milks).

Further Reading (starter, non-regionalized)

• Textbook chapters on human nutrition (macro/micro fundamentals).

• National dietary guidelines relevant to your target population (for local thresholds).

• Method papers on dietary pattern scores and nutrient density indices.

Takeaway

For AI, food must become structured, comparable, and contextual. Express meals in grams, kcal, %kcal, attach quality and pattern features, and preserve uncertainty flags. With that foundation, your models can reason about satiety, adherence, and risk—and your planners can deliver culturally fitting, safe, and achievable recommendations.