Exhaustive Academic Notes on AI Scientific Agents, Quantitative Genetics, Crop Breeding Pipelines, and Repository Workflows

Course Infrastructure, Team Repositories, and Agent Workflows

• Guest Speaker and Platform Access:
  • Aryan Thakur from Tube.ai will deliver a $30\text{-minute}$ presentation on AI agents for scientific workflows and building "AI scientists".
  • Platform access to Tube.ai's beta version will be provided to allow hands-on testing and feedback.
• Technical Readiness Checks:
  • Low-stakes presentations termed "technical checks" will evaluate student progress.
  • Students will interact with Claude Pro within assigned agent repositories to complete these technical checks.
• Computational Ag Lab Team Structure:
  • The lab is organized into three specialized commodity teams:
  • Viper Cotton Team
  • Viper Peanut Team
  • Viper Poultry Team
  • An instructor-led Corn Agent repository serves as a baseline model addressing agricultural challenge questions for the year $2050$
  • The hyperag_agents repository serves as a centralized hub for shared agent templates and baseline code.
• Repository Access Permissions and Forensics Protocols:
  • Write Access: Assigned exclusively to a team's dedicated repository (e.g., the Cotton team holds write permissions only for the Cotton repository).
  • Read Access: Granted across all team repositories to encourage cross-repository skill sharing.
  • Skill Attribution Rules: Any code, tool, or skill imported from another team's repository must contain explicit reference citations to its origin repository.
  • Repository Forensics: Claude-based forensic checks will be executed to audit code origins and confirm proper attribution.
• Claude Pro Account Usage Protocols:
  • Each commodity team receives dedicated Gmail accounts linked to Claude Pro accounts.
  • Usage allocation consists of recurring $5\text{-hour}$ operational usage windows per user.
  • Standard Target: Each team member should complete at least one full $5\text{-hour}$ development session per week.

Fundamentals of Plant and Animal Breeding, Genetics, and Domestication

• Core Definition of Breeding:
  • The intentional selection and controlled mating of plants or animals to increase the frequency of desirable traits in future generations.
  • Breeding relies on accumulating incremental genetic changes across successive generations.
• Historical Evolution of Domestication:
  • Canine Domestication: Wolves were domesticated into human-tolerant dogs prior to any agricultural plant or food animal domestication event.
  • Agricultural Domestication: Crop and livestock domestication began approximately $10,000$ to $11,000\text{ years}$ ago.
  • Maize Domestication: Occurred approximately $9,000\text{ years}$ ago in Central America, transforming wild Teosinte into modern corn (Zea mays).
  • Scientific Breeding Era: Formalized during the 1700s and 1800s via breeding societies for organisms such as flowers and pigeons. Gregor Mendel's genetic laws (established in the late 1800s and rediscovered in $1900$) enabled systematic pedigree tracking across parents, grandparents, and offspring.
• Morphological Trait Transitions in Domestication:
  • Teosinte vs. Modern Corn:
  • Wild Teosinte features a bushy weed-like structure with heavy tillering, highly shattering seed heads (seeds drop naturally to the ground), and seeds enclosed in hard, indigestible fruitcases.
  • Domesticated corn features a single erect central stalk, non-shattering ears (seeds remain attached for easy harvesting), unencapsulated naked kernels, and a significant increase in total seed count per ear.
  • Tomatoes: Transitioned from small, toxic wild berries into large, sweet, edible fruits.
  • Livestock: Evolved from wild boars into domestic swine, and from wild jungle fowl into highly specialized domestic poultry breeds.
• The Three Core Pillars of Genetic Improvement:
  • 1. Genetic Variation: Phenotypic differences within a population tied directly to underlying genetic diversity.
  • 2. Selection: Choosing superior individuals from a population mean to serve as parents, thereby shifting the trait mean of the progeny generation in a favorable direction.
  • 3. Inheritance/Heritability: Phenotypic traits passed from parents to offspring via Mendelian inheritance. Offspring inherit approximately $50\%$ of nuclear DNA from the maternal parent and $50\%$ from the paternal parent, modified by meiotic recombination shuffling grandparental alleles.
• Genotype-by-Environment Interaction ($G \times E$):
  • A specific genotype ($G$) performs differently across varying environmental conditions ($E$), such as drought, high ambient heat, nutrient availability, or latitude shifts.
• Intraspecific Diversification Examples:
  • Brassica rapa: Cabbage varieties diversified via selection for different vegetative traits.
  • Brassica oleracea: A single plant species selectively bred into distinct cultivars including kale, broccoli, cauliflower, cabbage, and Brussels sprouts.

Comparative Plant vs. Animal Breeding Systems and Quantitative Equations

• Biological and Operational Differences:
  • Plant Systems:
  • Stationary organisms, enabling precise field phenotyping.
  • High progeny yield per mating event (e.g., $300$ to $600$ seeds per single corn cross).
  • Capacity for self-pollination (inbreeding) to fix homozygous lines.
  • Facile clonal propagation (e.g., potato tubers allow $100\%$ genetically identical clones to be evaluated across multiple physical locations and years).
  • Animal Systems:
  • Mobile organisms requiring controlled facilities.
  • Low progeny yield per mating and extended generation intervals (e.g., cattle require $2\text{ years}$ to $3\text{ years}$ per cycle, whereas annual crops require $3\text{ months}$ to $4\text{ months}$).
  • Inability to self-pollinate; cloning is technically complex.
  • Deep historical pedigree and performance records spanning decades across extensive familial networks (parents, grandparents, half-sibs, aunts, uncles).
  • Operations bounded strictly by animal welfare standards.
• Classic Multi-Year Selection Pipeline:
  • Step 1: Parental Crossing: Select complementary parents and cross-pollinate to create new recombinants ($\approx 1,000,000$ untested progeny).
  • Step 2: Multi-Stage Testing: Evaluate recombinants in multi-year field trials. Reduce population size by $1$ to $2$ orders of magnitude each year while increasing the number of testing environments and field replicates.
  • Step 3: Variety Release: Select the top $1$ to $10$ elite lines for commercial release out of the initial $1,000,000$ candidates.
• Breeding Gain Equation and Genomic Selection:
  • Traditional selection cycles ($L$) require $8\text{ years}$ to $12\text{ years}$ to complete field performance testing.
  • The response to selection per unit time is defined by the breeding equation:     $\Delta G = \frac{H^2 \times I \times \sigma_g}{L}$     Where:
  • $\Delta G$ = Genetic gain per year
  • $H^2$ = Heritability of the target trait
  • $I$ = Selection intensity applied to the population
  • $\sigma_g$ = Additive genetic variance within the population
  • $L$ = Generation interval or parental recycling time in years
  • Genomic Selection Technology: Uses genome-wide DNA markers to predict breeding values based on training populations, shortening parental cycle time ($L$) from years down to months by bypassing multi-year field trials.

Agronomic Drivers, Historical Yield Trends, and 2050 Climate Targets

• Historical Trajectory of US Corn Yields ($1860$ to $2020$):
  • $1860$–$1940$: Yields remained static at approximately $20\text{ bushels/acre}$. Selection was limited to informal mass selection of open-pollinated varieties by individual farmers.
  • $1930s$: Geneticist George Shull discovered heterosis (hybrid vigor) by crossing inbred parent lines, establishing hybrid corn technology and founding the journal Genetics.
  • $1940$–$2020$: Commercial hybrid adoption reached $50\%$ in the $1940s$/$1950s$ and $100\%$ by $1960$. Average yields rose from $20\text{ bushels/acre}$ to $180\text{ bushels/acre}$–$200\text{ bushels/acre}$. Total planted acreage remained equal to $1930$ levels while total production output increased $7\times$
  • Cold War Geopolitics: Soviet Premier Nikita Khrushchev visited Iowa corn fields to study hybrid yield gains, coining the term "sausage on a stalk".
• 2050 Agricultural Production Challenges:
  • Global food production must increase by $56\%$ to sustain an estimated population of $1.0 \times 10^{10}\text{ people}$ by the year $2050$
  • In $2022$, severe drought impacted $14\%$ of global crop production acreage.
  • Tar Spot (Phyllachora maydis): A tropical fungal disease identified in Chicago in $2015$ that rapidly spread across uniform corn belt genetics. Causes yield losses of $20\text{ bushels/acre}$ to $30\text{ bushels/acre}$ ($10\%$ to $20\%$ total yield loss). Resistant tropical germplasm causes a $50\%$ yield drag when crossed into commercial elite lines.
  • Relative Maturity (RM) Bands: Range from $70\text{-day}$ maturity corn in northern latitudes (e.g., Canada) to $120\text{-day}$–$125\text{-day}$ corn in the southern US. Maturity must match local solar radiation and growing season length.
  • Short Corn Architectures: Developed in response to severe weather events like the $2020$ Iowa Derecho, which caused $2.0 \times 10^9\text{ USD}$ in crop damage. Short corn reduces height from $12\text{ ft}$ to $7\text{ ft}$ with thicker stalks.
  • Engineering Trade-Offs: Reduced ear height causes ears to fall below combine header harvest thresholds; denser stalk biomass slows combine operating speeds.
  • Poultry Broiler Mortality: Premature broiler mortality increased from $3.6\%$ in $2012$ to $6.0\%$ currently, generating significant feed conversion inefficiencies.

Agent Control Architecture, Repository Protocols, and GitHub Workflows

• Agent Operational Principles:
  • An agent requires a defined job, explicit metrics for success/failure, structured workflows, and output verification mechanisms.
  • Human-in-the-Loop Safeguards: Agents must explicitly signal uncertainty, ask clarifying questions, and obtain user permission before writing downstream files to prevent repository corruption.
• Development Environment Setup:
  • Integrated Visual Studio Code (VS Code) utilizing the Claude Extension.
  • Compute Scaling Roadmap: Local execution -> Local Workstations (for large language genome editing models) -> High-compute $H100$ GPU clusters by weeks $8$–$9$
• Repository Directory Layout:
  • CLAUDE.md: Primary persistent context and control file containing project boundaries, domain rules, operational protocols, and data access classifications.
  • Subdirectories: data/, docs/, evaluation/, knowledge/, notebooks/, skills/, tools/.
  • Contribution Notebooks: Individual contribution statements stored in notebooks/ as FirstName_LastName_Weekly_Contribution_WeekNumber.md.
• Data Access Permissions:
  • Repositories enforce Tier A Open Access Data rules to ensure all scraped external data can be legally re-published and used for model training.
• Team GitHub Handoff Protocol:
  • Step 1: Execute git pull before beginning any development session.
  • Step 2: Require Claude to present an explicit step-by-step plan before executing terminal commands or scripts.
  • Step 3: Perform execution, verify outputs, generate artifacts, and execute frequent git commit and git push commands.
  • Step 4: Notify teammates upon ending a session so the next user can pull the updated repository.
• Weekly Individual Contribution Reporting Schema:
  • 1. Planned tasks versus completed tasks.
  • 2. Concrete verification evidence (Git commit logs, line insertion/deletion metrics).
  • 3. Peer review feedback provided to other team members.
  • 4. Verified learning points.
  • 5. Active blockers and next planned steps.

Practical Demonstration: Soil Data Pipelines and Genomic Agent Modeling

• Driving Demonstration Question:
  • How to design corn crops for $2050$ in the US under shifting climate maturity bands and variable soil conditions.
• Soil Data Pipeline Implementation:
  • Integrated USDA SSURGO (Soil Survey Geographic Database) Tier A open data.
  • Extracted variables: Available Water Content (AWC), Organic Matter (OM), and Composite Soil Index.
  • Findings: Soil organic matter declines systematically moving south across US latitudes.
  • Unsupervised K-Means Clustering: Grouped US counties and survey regions into environmental clusters to evaluate crop performance across distinct soil zones (e.g., glaciated soils in Iowa).
• Genomic Selection Agent Evaluation and Failures:
  • The agent generated realistic-looking genetic gain curves ($\Delta G$ over selection cycles), but hallucinated backend execution rather than running quantitative genetic algorithms.
  • Simulated crosses between $110\text{-RM}$ / $115\text{-RM}$ and $120\text{-RM}$ lines yielded static single-point outputs without reflecting offspring recombination variance across local versus foreign soil types.
  • Takeaway: Prompting alone is insufficient; agents require rigorous programmatic backend validation tools, custom explicit skills, and compact context management (/compact) to prevent performance degradation.

Technical Check Requirements and Session Logistics

• Technical Check 1 Presentation Requirements (15 Minutes Total):
  • Slide 1: Baseline performance of an unassisted LLM on the team's core driving question (without custom context or tools).
  • Middle Slides: Current repository agent capabilities, pipeline executions, soil/genomic data structures, and generated artifacts.
  • Final Slides: Individual weekly contribution statement slides for each team member.
• Claude Extension Execution Modes:
  • Model Selection: Claude 3.5 / Claude 3 Opus.
  • Context Management: Execute /compact to compress conversation history and conserve token allowances.
  • Execution Modes: Manual, Edit, Plan, and Auto (Auto grants full permissions for terminal execution, local code runs, GPU/CPU access, and web scraping).
• Upcoming Logistics:
  • Aryan Thakur presenting next week on the Tube.ai platform.
  • Instructors attending virtually while Aryan Thakur presents in-person; joint session streaming with the Tifton campus.