RAD-Seq Project Planning for 384–600 Samples: DNA Input, Plate Batching, and Data QC

Figure 1: A 384–600-sample RAD-seq project functions as one linked design across several plates, from cohort definition and DNA intake to locus recovery, batch review, and analysis-ready delivery.

A 384–600-sample RAD-seq project should be planned as one biological study implemented across several linked plates, not as a series of independent 96-sample batches. The design must align restriction strategy, DNA quality, plate balance, technical controls, pilot evidence, locus consistency, missing-data rules, and population-genetic deliverables before production begins. This guide focuses on the decisions needed for a quotation and an auditable execution plan rather than repeating a general comparison of RAD technologies.

Key takeaways

• Define the biological comparison, sample roles, population structure, ploidy, reference availability, and primary analysis before choosing the restriction design.
• Review DNA as a cohort distribution; a few unusually concentrated samples do not compensate for systematic degradation or inhibitors in one group.
• Distribute major biological and extraction variables across plates so that population, phenotype, location, family, or tissue is not identical to a processing batch.
• Use a representative pilot to test locus recovery, fragment consistency, per-sample balance, missingness, and downstream usefulness before all libraries are committed.
• Connect each QC metric to an action such as release, re-pooling, added sequencing, repeat library preparation, re-extraction, or documented exclusion.

Scope the Cohort Before Plating

The first planning decision is not how many wells are available. It is what comparison the study must support. A diversity survey, population-structure analysis, linkage population, selection scan, parentage study, and marker-discovery project can all use RAD-derived data, but they place different demands on sample representation, marker density, missingness, relatedness, and reference-based analysis.

For 384–600 samples, small ambiguities multiply. A sample count without group labels cannot show whether populations are represented across plates, and “SNP calling” does not define whether the primary output is a raw VCF, filtered matrix, linkage dataset, or PCA-ready file. Freeze the biological unit and intended inference before barcodes are assigned.

Researchers who still need to choose among original RAD, ddRAD, GBS, and 2b-RAD can use the RAD-seq technology comparison before moving into production planning. Once a RAD-family approach is selected, the RAD-Seq service provides the method-specific intake context that should be reconciled with the study design.

Fix the Restriction Design

Restriction design defines which fraction of the genome can become observable. Enzyme recognition frequency, methylation sensitivity, genome composition, repeat content, target fragment window, and reference availability all affect the expected locus set. A design that produces too many fragments may distribute reads too thinly, while one that produces too few may not provide enough informative markers for the intended population-genetic question.

When a suitable reference assembly exists, in-silico digestion can estimate the number and genomic distribution of candidate fragments under alternative enzyme combinations and size windows. The estimate is not a guarantee: assembly gaps, structural variation, methylation, polymorphism at restriction sites, and wet-laboratory size selection change what is recovered. Without a reference, closely related genomes and a pilot library can still narrow the design, but uncertainty should be stated explicitly.

The quotation should record:

• the RAD variant and restriction enzyme combination under consideration;
• whether enzyme choice is established for the species or requires evaluation;
• the planned fragment-size window and how consistently it can be reproduced across libraries;
• whether PCR duplicate identification is available in the selected design;
• expected reference-based or de novo locus processing;
• the downstream analysis that the locus set must support.

The ddRAD-seq explainer describes why dual digestion and size selection affect locus recovery. Projects that require a tunable fragment subset can then review ddRAD-seq services. Where fixed-length type IIB tags fit the biological question, the 2b-RAD plant genomics guide and 2b-RAD services provide a separate design route. These pages support method choice; the present guide addresses what happens after that choice enters a multi-plate cohort.

Treat DNA as a Cohort Variable

RAD library preparation begins with enzymatic digestion, so DNA amount alone is not a sufficient acceptance criterion. Integrity, purity, solvent, extraction chemistry, freeze-thaw history, and matrix-derived inhibitors can alter digestion and representation. Fluorometric concentration estimates usable DNA mass, whereas absorbance ratios help flag contaminants; neither measurement replaces an integrity check.

Review the cohort distribution before deciding which samples can be normalized together. If all samples from one field site were extracted with a different method and show lower integrity, merely assigning them random well positions does not remove the underlying difference. They may require cleanup, re-extraction, a separate pilot stratum, or a documented interpretation limit.

Normalization should produce comparable inputs without consuming the only aliquot of scarce material. Define the reserve amount, repeated intake measurements, and whether borderline samples enter the pilot. A rescue policy should distinguish re-quantification, cleanup, re-extraction, repeat library preparation, and biological replacement because these actions solve different problems.

Build Plates Around Biology

Four 96-well plates provide 384 physical positions, not room for 384 study samples plus controls and repeats. Six hundred study samples require at least seven plates before additional control, repeat and reserved positions are counted; more plates may be needed. Calculate plate count from the complete well-allocation manifest. The harder task is distributing biological groups and technical variables so that plate cannot imitate the comparison of interest.

Use constrained randomization. Each plate should represent the relevant populations, families, locations, phenotypes, sample matrices, and extraction batches. Pure shuffling may break family structure or leave rare groups uneven, so audit balance from metadata rather than appearance.

Figure 2: A constrained plate map distributes the principal biological groups and extraction histories while carrying defined controls across the multi-plate project.

Preserve sample identity

The sample manifest should use one immutable study ID and separate fields for source ID, extraction ID, plate, well, barcode, library, sequencing run, repeat number, and final analysis ID. Do not encode all meaning in a long sample name or rely on plate color alone. Machine-readable fields make balance checks and later reconciliation possible.

Place controls with purpose

Each control should answer a stated question. Repeated reference DNA or selected duplicates can assess cross-plate comparability and identity. Negative controls can expose contamination or index-assignment issues. Parents or known relatives may support relationship checks, but they do not replace sample tracking.

Before release, inspect group counts by plate and cross-tabulate plate against extraction batch, population, phenotype, tissue, collection site, and operator. The broader GBS large-cohort planning guide discusses general plate balance; RAD projects add a further requirement to compare recovered loci and fragment behavior across libraries.

Pilot Locus Recovery First

A pilot is valuable when it resolves a decision, not when it is a miniature run of only the cleanest DNA. Include representative populations, extraction batches, matrices, quality extremes, ploidy states, and heterozygosity. Retain enough irreplaceable material to repeat informative samples.

The pilot should test the complete evidence chain from input QC through digestion, library construction, sequencing, locus recovery, and an initial genotype dataset. For multi-library designs, it should also indicate whether fragment-size distributions and retained loci overlap sufficiently for samples to be analyzed together.

Figure 3: Pilot evidence should lead to explicit scale-up, re-pooling, restriction redesign, sample remediation, or analysis-scope decisions.

Arantes and colleagues demonstrated the value of small-scale sequencing and weighted re-pooling in a large RADseq dataset. The exact thresholds from one study should not become universal acceptance rules, but the design principle is transferable: early evidence is most useful when it changes a production decision. Added reads can improve a shallow but otherwise sound library; they cannot repair failed digestion, contamination, a non-overlapping locus set, or biological groups perfectly confounded with plates.

Set Production QC Gates

Production QC should connect evidence to action. A list of read counts, mapping rates, missingness, and heterozygosity is descriptive; a release plan states what happens when a gate fails. Preliminary bands can be proposed before the pilot and finalized for the selected protocol, species, and objective.

Figure 4: An action-linked QC matrix distinguishes problems that may respond to additional sequencing from those that require laboratory or design review.

Review distributions, not only overall means. A project can have an acceptable average while one plate has fewer recovered loci or one population has higher missingness. Plot key metrics by every recorded batch variable and by biological group. Technical duplicates and shared controls are useful only when their expected concordance and review rule were defined before results were seen.

Interpret Missingness by Pattern

Missing genotypes in RAD data can arise from insufficient depth, unequal pooling, failed or partial digestion, sample degradation, inconsistent size selection, alignment or locus-clustering decisions, and polymorphism at restriction sites. These mechanisms do not have the same consequences. Restriction-site polymorphism can cause allele or locus dropout linked to ancestry, while shallow sequencing may create more stochastic missingness.

Overall missingness therefore needs a denominator and a pattern analysis. Report per-sample and per-locus missingness before and after filtering, then stratify by plate, library, run, population, family, tissue, extraction batch, and DNA-quality category. A modest group difference aligned with plate may be more concerning than a larger but randomly distributed rate.

Three distinctions guide interpretation:

• Random versus structured. Random loss mainly reduces information; group- or batch-aligned loss can bias population comparisons.
• Sample versus locus. A low-quality sample and a poorly represented locus call for different filters and repeat decisions.
• Technical versus biological. Depth imbalance may respond to re-pooling or sequencing, whereas restriction-site variation is part of the method's ascertainment process.

Sensitivity analysis should test whether PCA, relatedness, diversity summaries, ancestry estimates, or selection signals remain stable across defensible filter sets. Imputation, if used, must be reported separately from observed genotypes with method, reference data, software, and validation evidence. For projects that require more continuous genome-wide information, the low-pass WGS cohort planning guide provides an alternative design context rather than a promise that post hoc filtering can make the platforms equivalent.

Define Population-Genetic Deliverables

“VCF included” is not a complete delivery specification. A variant file may be raw or filtered, contain hard calls or genotype likelihoods, include multiallelic sites, or reflect imputation. The project plan should identify primary, intermediate, and analysis-ready files and connect each one to a named reference assembly, sample list, locus definition, and filter version.

A traceable package may include:

• demultiplexed FASTQ files and checksums;
• the final sample, barcode, plate, well, library, run, and repeat manifest;
• read-level and library-level QC summaries;
• reference, alignment, or de novo locus-assembly records as applicable;
• per-sample locus recovery, depth, missingness, and heterozygosity tables;
• raw and filtered VCF files with filter rules and pre/post-filter counts;
• observed and imputed datasets clearly separated when imputation is scoped;
• duplicate concordance, relatedness, PCA, or population-structure-ready formats where agreed;
• a batch-stratified QC report and exclusion or repeat log.

The agricultural genomic data analysis service can be scoped with RAD data generation when projects need variant processing and population-genetic summaries beyond raw sequencing. The analytical plan should still state what can be inferred from the final sample size and marker set; hundreds of samples do not rescue unbalanced populations, inadequate biological replication, or missing metadata.

Prepare a Quote-Ready Package

A useful quotation package lets the technical team evaluate feasibility without guessing what “600 RAD samples” means. Send a sample-level manifest and a short study brief rather than only a total count.

Include the following information:

• species, genome size if known, ploidy, reference assembly, and expected heterozygosity;
• primary research question and the one or two analyses that define success;
• total samples and counts by population, family, phenotype, location, generation, sex, or treatment;
• sample type, preservation, extraction method, extraction date, DNA QC evidence, buffer, and remaining volume;
• whether the restriction design is established, proposed, or needs pilot evaluation;
• known collection, extraction, plate, library, or seasonal batches;
• controls, technical duplicates, parents, founders, and reference samples;
• acceptable re-extraction, repeat, replacement, and exclusion rules;
• required raw, filtered, imputed, and analysis-ready deliverables.

Flag irreplaceable samples and identify the biological groups they represent. If a failed sample can be replaced only by another individual from a different population or family, that substitution changes the study rather than merely filling a well. A complete manifest also allows the plate design to be audited before physical shipment.

RAD-Seq Planning Support

CD Genomics supports agricultural and population-genetic RAD sequencing projects with consultation on method fit, sample and cohort review, restriction-design considerations, plate and control planning, pilot evaluation, library preparation, sequencing, and documented bioinformatics delivery. For a 384–600-sample project, the most useful first discussion combines the sample manifest, DNA QC distribution, biological grouping, reference resources, primary endpoint, and required deliverables. Exact intake and release criteria are confirmed for the selected RAD variant and project rather than copied from an unrelated study.

RAD-Seq Project Planning FAQ

References

1. Arantes LS, Caccavo JA, Sullivan JK, et al. Scaling-up RADseq methods for large datasets of non-invasive samples: Lessons for library construction and data preprocessing. Molecular Ecology Resources. 2025;25(5):e13859. doi:10.1111/1755-0998.13859
2. Chambers EA, Tarvin RD, Santos JC, et al. 2b or not 2b? 2bRAD is an effective alternative to ddRAD for phylogenomics. Ecology and Evolution. 2023;13(3):e9842. doi:10.1002/ece3.9842
3. Lou RN, Therkildsen NO. Batch effects in population genomic studies with low-coverage whole genome sequencing data: Causes, detection and mitigation. Molecular Ecology Resources. 2022;22(5):1678–1692. doi:10.1111/1755-0998.13559
4. Suchan T, Dufresnes C, Schmidt R, Ambu J, Gippner S, Vences M. ddRADseq for animal population genomics/phylogenomics. protocols.io. 2024;version 1. doi:10.17504/protocols.io.kxygx3nzwg8j/v1
5. Sethuraman A. The missing data problem in population genomics and statistical methods to address them. G3: Genes, Genomes, Genetics. 2026;16(1):jkaf269. doi:10.1093/g3journal/jkaf269
6. Andrews KR, Good JM, Miller MR, Luikart G, Hohenlohe PA. Harnessing the power of RADseq for ecological and evolutionary genomics. Nature Reviews Genetics. 2016;17(2):81–92. doi:10.1038/nrg.2015.28
7. Peterson BK, Weber JN, Kay EH, Fisher HS, Hoekstra HE. Double digest RADseq: an inexpensive method for de novo SNP discovery and genotyping in model and non-model species. PLOS ONE. 2012;7(5):e37135. doi:10.1371/journal.pone.0037135

Research Use Only Statement

This content and the described services are intended for agricultural, ecological, evolutionary, and population-genetic research. They do not provide clinical diagnosis, treatment decisions, or individual health assessment.