Attachment 3 - Statement of Work - 1232SA26Q0936.pdf

USDA/ARS Statement of Work (SOW)

Project Title: Development of Atlantic salmon genomic breeding values

Requesting Office: USDA Agricultural Research Service (ARS), National Cold Water Marine Aquaculture Center

Program/Project: Improving North American Atlantic Salmon Aquaculture Genetics

Period of Performance:

Three (3) year ordering period.

Place of Performance: Virtual/Contractors Place of Business

1. Background

USDA ARS conducts genetic improvement programs to enhance performance and resilience in aquaculture species. Reliable genomic data processing, and robust prediction of genomic estimated breeding values (gEBVs) for economically important traits are essential to achieving genetic gain while controlling inbreeding and preserving genetic diversity. This SOW engages a qualified contractor to deliver end-to-end genomic QC, and gEBV computation for Atlantic salmon maintained by the National Cold Water Marine Aquaculture Center and other related genomic services. The exact yearly scope of work could be smaller based on government needs as the contract progresses.

2. Objectives

1. Genotype Data QC: Clean, validate, and document genotype data to an agreed standard suitable for downstream analysis. Assign parentage to fish based on genomic data and quality control pedigree file. Inconsistencies between genomic data and pedigree file will be documented and recommendations for how to resolve the conflict suggested to USDA for a final decision.

2. gEBV Estimation: Generate growth and sea-lice resistance specific gEBVs (and accuracies/reliabilities) for all eligible individuals in the breeding population. The gEBVs will be transformed into a selection index to weight traits according to USDA determined values.

3. Quantify genetic parameters and conduct GWAS for fillet fat content and pigment:

Utilize provided phenotypic data to assess additive genetic and residual variance-covariance components for fillet fat content and pigment. Determine independent trait heritability values and genetic covariation among traits and conduct a genome-wide association study (GWAS) to characterize the genomic architecture of traits by identifying specific loci and genomic regions associated with traits of interest.

4. Mate Allocation: Identify mating pairs that meet genetic gain objectives while controlling mean co-ancestry/inbreeding and operational constraints. Males can be used to mate with a maximum of two individual females. Alternative males will be identified for each female.

5. Database Management: Assist in transferring all data developed or used as part of the project into the USDA’s choice of database at the end of each year.

3. Scope of Work & Tasks

Task 1: Genotype Data Quality Control

The contractor shall:

• Data Intake & Inventory o Accept genotype data in formats including VCF, PLINK (either .ped and .map or .bed, .bim, and .fam), .cel format (CEL files), SNP array reports, and/or imputed datasets.

This may include genotype data from multiple generations.

o Accept phenotype data files in excel; metadata includes sample IDs, sex, population/strain, collection date, and laboratory batch. This may include phenotype data from multiple generations.

o Accept pedigree files in excel.

o Crosswalk sample identifiers to USDA master roster; report any mismatches, duplicates, or missing samples.

• QC Criteria & Processing o Apply sample-level filtering: call rate (≥ 95%), heterozygosity outlier detection, sex checks (using sdY markers from the SNP array), relatedness checks for duplicates/contaminants.

o Apply marker-level filtering: genotype call rate (≥ 95%), minor allele frequency (MAF ≥ 0.05), Mendelian incompatibility, and with departures from Hardy–Weinberg equilibrium (HWE) lower than 0.15, based on the difference between expected and observed frequency of heterozygotes (|Het expected – Het observed| < 0.15) to reduce technical artifacts and remove monomorphic/technical artifacts.

o Batch effects assessment, plate effects, and population stratification checks (e.g., PCA).

o Document all pipeline steps, software versions (e.g., PLINK, bcftools), parameters, and thresholds.

• QC Outputs o Clean genotype dataset suitable for downstream analysis.

o QC report: summary tables, figures (call rates, MAF spectra), lists of excluded samples/markers and reasons for the exclusions.

o Reproducible scripts/workflows (using bash/R/Python or similar) o Data dictionary and SOP.

• Pedigree File QC o Utilize the genotyping data to quality control the provided pedigree file data.

o Assign parentage based on genotypes to validate the pedigree records and/or detect possible errors in samples processing from fin tissue collections to DNA extractions to genotyping with the 50K SNP array.

o Compare genomic parentage assignments with provided pedigree, document any inconsistencies and suggest plans to reconcile the two. All final decisions on how to reconcile any inconsistencies between the genomic parentages and pedigree file rest with the USDA.

o Document all pedigree inconsistencies and resolutions.

o Filter out all animals with >1% SNP genotypes mismatch to their expected parents based on the pedigree records.

Task 2: GEBV Estimation for Defined Traits

The contractor shall:

• Modeling Framework o Use appropriate genomic prediction methods: ssGBLUP (single-step), following

USDA bioinformatic guidelines and published literature. Other methods can be used if pre-approved by USDA.

o The prediction model has to be developed in consultation with and approved by USDA and may include fixed effects (e.g., hatchery/tank, batch, age, sex) and relevant covariates; specify random effects.

• Traits & Data o Traits: growth rate, sea lice resistance.

o Phenotypes: harvest evaluation data, sea lice infection data; perform phenotype QC

(outliers, transformations if needed).

o Pedigree: available; utilize quality-controlled pedigree file with approved reconciliations for identified inconsistencies.

• Validation & Metrics o For each trait provide estimates of heritability and genetic correlation between current year-class and parental year-class, if phenotype data available from the parental year-class. If genotypes are available from the parental year-class in addition to phenotypes, then also provide heritability and genetic correlation estimates using the genomic information.

o Report correlation between pedigree-based EBV and genomic-based GEBV, and between GEBVs from genomic predictions models, if more than one model is used.

o Report estimates of accuracy, mean squared error (MSE), area under the curve (AUC) and bias from cross-validations of pedigree-based EBV (PBLUP) and genomic-based GEBV to assess predictions’ performance of evaluated prediction models (e.g., PBLUP, and GBLUP, or ssGBLUP, or wssGBLUP, or Bayesian BayesB, etc.).

• GEBV/EBV Outputs o Per-individual EBV and GEBV (and reliability or bias estimate) for all traits.

o Summary report: methodology, parameterization, diagnostics, convergence checks.

o Reproducible code and configuration files (e.g., BLUPF90/ASReml/GCTA/BOLT-

LMM, etc.), including version number for each software package.

Task 3: Quantify genetic parameters and conduct GWAS for fillet fat content and pigment:

The contractor shall:

• Data Intake & Inventory o Accept genotype data in formats including VCF, PLINK (either .ped and .map or .bed, .bim, and .fam), .cel format (CEL files), SNP array reports, and/or imputed datasets;

metadata includes sample IDs, sex, population/strain, collection date, and laboratory batch.

o Pedigree files in excel.

o Accept phenotype data in excel for: EPD, DHA, and astaxanthin (per 100g of fillet), total fat content (% of tissue weight), total Omega-3 fatty acids per serving, and carcass weight (kg at harvest).

o Crosswalk sample identifiers to USDA master roster; report any mismatches, duplicates, or missing samples.

• Modeling Framework o Use appropriate prediction methods: BLUP and ssGBLUP (single-step) to estimate additive genetic and residual variance-covariance components for various fat content and pigment traits with and without genomic information.

o Conduct genome-wide association study (GWAS) using ssGBLUP, or other methods if pre-approved by USDA, to characterize the genetic architecture of the traits by identifying quantitative trait loci (QTLs) and genomic regions associated with the traits.

o Estimate the proportion of genetic variance explained by each QTL window, and each individual SNP within the QTL windows.

o Identify candidate causal genes located close to the identified loci following practices established in the scientific literature o Summary report: methodology, parameterization, diagnostics, convergence checks (if applicable), and results.

o Reproducible code and configuration files (e.g., BLUPF90/ASReml/GCTA/BOLT- LMM, etc.), including version number for each software package.

Task 4: Mating Pair Identification (Mate Allocation)

The contractor shall:

• Objective & Constraints o Optimize mate allocation to maximize genetic gain subject to constraints on expected inbreeding, mean co-ancestry, and effective population size (Ne).

o Enforce operational constraints: sex ratios, tank capacity, number of families, minimum/maximum matings per parent, availability windows, quarantine/health status, and logistics as specified by USDA .

o Avoid mating of close relatives (e.g., kinship threshold ≤ 0.0625–0.125, configurable upon approval from USDA).

• Methodology o Use Optimal Contribution Selection (OCS) and/or constrained mate allocation based on genomic or pedigree-based kinship matrices or other method if approved by USDA.

o Incorporate weighted multi-trait selection index (provided by USDA) or single-trait prioritized lists if directed to use by USDA instead of the weighted index.

o Provide sensitivity analysis under alternative constraints (e.g., different inbreeding ceilings).

• Mate Allocation Outputs o Final mating plan: pairings, expected inbreeding coefficients per family, projected genetic gain, contribution per parent.

o Alternative selection pressure scenarios (e.g., conservative, balanced, aggressive gain).

Implementation notes for hatchery operations (batching, tagging/ID, contingency plans).

Task 5: Database Management

The contractor shall:

• Objective & Constraints o Create a list of all data generated, used or complied as part of the contract.

o Format all the data above in a manor to be easily uploaded into a database of the

USDA’s choice.

o Assist with exporting data to USDA and integration of project data into the database of USDA’s choice.

Continuation of Annual Activities:

If option years are exercised, a set of activities from those above will be defined in the scope of work for the option year.

4. Government-Furnished Information/Materials (GFI/GFM)

USDA/ARS will provide:

• Raw genotype files .cel format (CEL files) (VCF, PLINK, SNP array reports) and metadata.

• Phenotype databases and data definitions for specified traits.

• Pedigree files.

• Previous year analyses [If desired] for continuity.

5. Contractor Qualifications

• Demonstrated expertise in aquaculture genetics, genomic data processing, and GEBV estimation.

• Experience with relevant software (PLINK, BLUPF90 suite, ASReml, GCTA, BOLT-LMM, AlphaMate/Custom OCS tools).

• Capacity to deliver reproducible, auditable workflows; familiarity with USDA/ARS data stewardship and federal IT security practices.

• [Preferred] Experience with Atlantic salmon and traits analogous to those in scope.

• Company shall have molecular laboratories and personnel employed and physically located in the United States at the time of due date of this RFP.

6. Deliverables & Formats

Deliverable Description Format Due

D1: Project Plan Work breakdown, schedule, risk register, data handling plan PDF + MS Word

July 1 of current FY (or 2 weeks after contract is awarded; whichever is later)

Deliverable Description Format Due

D2: QC Report & Clean Genotypes

Full QC outputs, cleaned datasets, scripts, SOP

PDF report;

PLINK/VCF;

scripts in report

August 1

D3: GEBV

Results

GEBVs, reliabilities, model specs, validation metrics

Excel + PDF report; code August 15

D4: Individual Mating Plan

Primary and alternate matings for each eligible female determined on a weekly or semi-weekly basis

Excel 18 hours after list of eligible males and females sent

D5: Fillet Fat and Pigment GWAS

Consolidated methods, results of Task 3 work, including full scripts, codes and markers

PDF + Word;

scripts in report;

code

60 days before end of

POP

D6: Final Technical Report

Consolidated methods, results, recommendations PDF + Word End of POP

D7: Knowledge Transfer Virtual briefing; training materials PPT + recorded session As scheduled in project plan

D8: Information Transfer

Delivery of data for integration into USDA Database TBD Within 60 days of the end of spawning

Data & Code Packaging:

• Use non-proprietary formats where feasible; USDA approval if proprietary formats are to be used.

• Ensure version history for data analyses/manipulations is preserved for auditability.

• Document all changes to the pedigree file made, including why the change was made and

USDA representative that authorized the change

Deliverables:

For each deliverabler, the contractor shall provide updated versions of Deliverables D2 (QC Report & Clean Genotypes), D3 (GEBV Results), D8 (Information Transfer) and any additional analyses specified in the Option Year scope of work. Deliverables are due on the same timeline relative to the start of each option year unless modified in writing by the Government.

7. Performance Standards & Acceptance Criteria

• QC Completeness:

o Apply sample-level filtering: call rate (target ≥ 95%), heterozygosity outlier detection, sex checks (using sdY markers from the SNP array), relatedness checks for duplicates/contaminants.

o Apply marker-level filtering: genotype call rate (≥ 97%), minor allele frequency (MAF ≥ 0.05), Mendelian incompatibility, and Hardy–Weinberg equilibrium (HWE) within expected Het frequency |Het expected – Het observed| < 0.15| to reduce technical artifacts, removal of monomorphic/technical artifacts.

o Clear documentation of exclusions and rationale; Pedigree check with genotype data and relatedness diagnostics included.

• Model Transparency & Reproducibility:

o Full disclosure of model formulations, priors (if Bayesian), convergence diagnostics, and software versions.

o Pipelines reproduce USDA verification results within tolerance (e.g., correlation differences ≤ 0.01).

• GEBV Quality:

o Predictions’ reliability estimates provided for each animal and averaged for the population o For each trait provide estimates of heritability and genetic correlation between current year-class and parental year-class, if phenotype data available from the parental year-class. If genotypes are available from the parental year-class in addition to phenotypes, then also provide heritability and genetic correlation estimates using the genomic information.

o Report correlation between pedigree-based EBV and genomic-based GEBV, and between GEBVs from genomic predictions models, if more than one model is used.

o Report estimates of accuracy, AUC, MSE and bias from cross-validations of pedigree-based EBV (PBLUP) and genomic-based GEBV (ssGBLUP) to assess performance of prediction models. (and to all other genomic prediction models used)

• Mate Allocation Quality:

o Expected mean co-ancestry/inbreeding ≤ agreed ceilings.

o Operational constraints satisfied; sensitivity scenarios provided.

• Timeliness:

o 100% of milestones met by scheduled dates absent approved change requests.

USDA/ARS will review deliverables within 15 business days and either accept, request clarifications, or require revisions.

8. Security, Privacy, and Compliance

• Data Classification: [non-CUI]; contractor shall handle data per USDA and federal requirements.

• IT Security: Comply with applicable FISMA controls aligned to Moderate baseline when accessing USDA data systems; use encrypted transfer/storage (e.g., SFTP, HTTPS, AES-256 at rest).

• Access Control: Named personnel only; MFA; least privilege.

• Data Rights & IP: See Section 9.

9. Data Rights, Ownership, and Publication

• Ownership: All data, and reports generated under this contract are USDA property.

Proprietary analyses and code pipelines shall remain the property of the contractor under the condition that they will be made available to the USDA for review.

• License: Contractor grants USDA a perpetual, royalty free, worldwide license to use, modify, and distribute deliverables for government purposes.

• Publication: Contractor may not publish or publicly disclose results without prior written USDA approval.

• Confidentiality: No disclosure of sensitive data to third parties; NDAs for subcontractors as needed.

• Data Use: Contractor may not use USDA data either provided by USDA or derived from this project in any format for other clients. This includes using models trained with USDA data to analyze data from other clients.

10. Reporting Requirements

• Kickoff Meeting: Within 10 business days of award.

• Biweekly Status Reports: Progress, issues, risk updates, upcoming milestones.

• Quarterly Technical Briefings: Methods deep dives and interim findings.

• Closeout Meeting: Presentation of final results and handover.

11. Standards & References

• Follow recognized best practices in genomic QC, mixed models/BLUP, and optimal contribution selection.

• Adhere to USDA/ARS Data Management and IT Security policies, and applicable federal standards (e.g., NIST SP 800 series for security controls).

• Use species-appropriate parameters; document any deviations and justifications.