Independent quantitative genetics and bioinformatics consulting
Know which lines to advance, which genetic signals are worth following up, and how far you can trust your prediction models, from the data you already have. For breeding teams, ag-biotech companies and research groups: fixed scope and price, results in weeks, and the code stays with you.
What I solve
Start from the decision, not the method. You get the analysis your question, trial design and data call for, and you pay for nothing you do not need.
Selection
Trial data from several sites and seasons, and a selection meeting on the calendar.
You get a ranked list of candidates, how stable each one is across environments, and how much of the difference between them is genetic, so you advance lines on genetic merit, not on one good season.
Genetic signals
A GWAS has produced a list of hits, or you have the data and no time to run one.
You get the associations that hold up, the ones that do not, and where your validation budget is best spent.
Genomic prediction
Genomic prediction is about to decide which lines never reach the field.
You get its predictive ability measured the way you will use it, on new families, new environments or the next cycle, so you know whether to trust it before good lines are dropped.
For research groups: defined analysis questions answered, grant reports delivered and manuscripts finished.
Why work with me
You get the statistics and the bioinformatics from someone who knows what a breeding team does with the result: answers you can use in your next selection round.
Trained in genetics and plant breeding, I build the analysis around your trial design, traits and selection goals. You receive decisions in your team's terms: which lines, which traits, and how much confidence to place in each.
Research exampleMaize: across 700+ hybrids and 34 traits, separated the traits worth selecting on from those a marker panel would miss.
Bioinformatics and statistical genetics often need two specialists, and each handover costs time and money. You have one contact from raw sequencing reads to the selection decision, and nothing gets lost in between.
Research examplePotato: 6,600+ accessions taken from raw reads to genotypes now used by three commercial breeding companies.
Mixed ploidy, species with no reference genome or breeding population, traits extracted from images: I have delivered results on each. If your data look like this, the project keeps moving.
Research examplesDiploid to hexaploid potato. The first reference genome of Kersting's groundnut. 24,000+ root-phenotyping observations reduced to the nine traits worth measuring.
A prediction is only worth acting on if it holds up where you will use it. I test models on new families and environments before you rely on them, and I have followed selections through to field trials.
Research exampleKersting's groundnut: index-based selection gains confirmed across field environments, +21% seed weight and −6% days to flowering.
Research record
Six projects from my research, from genebank genomics to field trials and phenotyping platforms. For each, the situations where the same approach would help you.
6,600+ potato accessions, from wild species to breeding lines and from diploid to hexaploid: a dataset that breaks standard pipelines. I lead all bioinformatics in the project: reliable genotypes from raw reads, used by the partners including three commercial breeding companies; a representative core collection; the training population for genomic prediction; a potato graph pangenome; and GWAS on nutrient-use efficiency. For a breeding company, that means genotypes you can trust in polyploid material and a training population chosen to cover your germplasm. For a genebank, pre-breeding or a trait that is costly to measure, a core collection that keeps the diversity while cutting the number of plants to phenotype.
Relevant ifyou hold a large or diverse collection, work with polyploid material, or have sequencing data that has not yet become usable genotypes.
The question: can landrace-derived hybrids be selected for biomass that converts more easily into bioenergy, and can genomic prediction speed that up? I analysed multi-environment trials of 700+ four-way hybrids, mapped 34 traits on 945k SNPs and tested genomic prediction, including on new families.
The answer: saccharification is polygenic, with many small additive effects; the two loci found explain at most about 7% of it. Under random folds, predictive ability reached up to 0.67 for grain traits but only 0.15 for saccharification; it fell on new families, and a second trial location added little.
What this saves a breeder or agri-tech company: spending on a marker panel or a few target loci for traits like this. Use genome-wide prediction, test it on new families before trusting it, and check what an extra trial location adds before paying for it.
Relevant ifyou run multi-environment trials and need to know which traits are worth selecting on, and whether genomic prediction will hold up on new material.
With almost no genomic resources to start from, I assembled the species' first reference genome (Nature Communications) and built its quantitative genetics: heritability across seven field environments, genomic prediction, and index-based selection whose gains were confirmed in the field, +21% seed weight and −6% days to flowering. Most of the DArTseq markers could not be confirmed by independent sequencing, yet the confirmed ones alone predicted as well or better for 11 of 15 traits. The practical step for your own marker set: a call-rate filter, which here recovered 97% of that gain without paying for a second genotyping platform. UAV-based phenotyping added image-derived field traits.
Relevant ifyou work on an underused crop or a population with little genomic history, or need to know whether your markers can be trusted.
Root systems shape how well a rootstock copes with low nitrogen, yet they are hard to measure. On a 378-plot rhizotron platform for the BOLERO project, 8 coffee species and 3 cocoa clones grown under two nitrogen levels produced 24,000+ observations across 137 variables. Of 56 image-derived root traits, 9 described the root system without redundancy; total root length kept a high heritability throughout the 66 days, while the heritability of the others shifted with plant age and nitrogen supply. Six genotypes developed a more efficient root system under low nitrogen, a first shortlist of rootstock candidates. The practical step for a phenotyping programme: score fewer traits, choose them for heritability, and fix when you measure them.
Relevant ifyour phenotyping platform produces more traits than you can interpret, and you need to know which ones are heritable and worth selecting on.
Across 222 sunflower genotypes measured with hand-held fluorescence instruments over two years and three developmental windows, heritability was highest in the latest window for 8 of 9 traits, and a season-wide average was less heritable than that window for all 9, at an average cost of 18.5% of selection response per cycle. Genomic prediction trained in one window did not carry over to another (median predictive ability 0.03, against 0.37 within a window). The right window depends on the crop and the trait, so it has to be set from your own data: that is the step I do for a client, identifying the window where your measurements are most heritable and training prediction models on it. A second study tested genomic prediction of the CO₂ compensation point across C3–C4 intermediate Helianthus species.
Relevant ifyou measure physiological traits during the season and need to know when to measure them and whether they can be selected on.
Across a 125-accession panel of Vigna unguiculata, a GWAS identified seven major QTLs (R² > 10%) for carotenoid content in sprouts. The practical step: screen candidate parents with these markers before crossing for biofortification.
Relevant ifyou want to turn a trait of interest into markers you can use to choose parents.
Core expertise
How it works
A short email is enough: the crop, the data you have and the decision you face. No commitment at this stage.
You approve the question, methods, deliverables and price in writing before any work starts. No open-ended bills.
Results you can act on: data checked, models validated, every result tied back to your decision.
Report and code in your hands, with a clear next step. Your team can rerun everything without me.
What you receive
Services and prices
Before any work starts, you receive a written scope: the question, the deliverable and the price. Your data stays confidential, under NDA if you wish, and the results and code are yours. Not sure which engagement fits? Start with the Diagnostic, or describe your dataset and I will suggest one.
| Offer | What you get | Fixed price |
|---|---|---|
| Data & Analysis Diagnostic | Know whether your data and plan can answer your question before you commit budget. A 60-minute working session on your data, code or analysis plan, then written recommendations within 48 hours. | 248 EUR |
| Pipeline Rescue | Get a broken or stalled R/Python pipeline running again, documented so your team can run it without its original author. Running pipeline, documentation, handover call. | 494 EUR |
| Analysis Sprint | A clear answer to one question on one dataset: GWAS, genomic prediction or phenotyping statistics, cross-validated. Results, figures, methods text and code, ready to act on or to report. | 755 EUR |
| Manuscript Rescue | Get a stalled paper moving again: your dataset and half-written draft taken to submission-ready analysis, with publication figures and methods and results sections. | 998 EUR |
| Core Collection & Training Set Design | Phenotype fewer plants without losing what matters: a representative subset chosen from your genotypes, as a core collection for a genebank or as the lines to phenotype so a genomic prediction model covers your germplasm. Subset list, representativeness report, code. | 750–1,500 EUR |
| Pipeline Build | Run your routine analyses in-house: a reproducible GWAS, genomic selection, GBS or phenotyping workflow that becomes yours. Scoped per proposal. | 2,000–5,000 EUR |
| Retainer Analyst | An analyst on call without a new hire: 5 to 10 hours per month for teams with a steady flow of data. Monthly summary, priority response. | 400–800 EUR/mo |
The Diagnostic fee is credited toward a Pipeline Build booked within 30 days.
Get in touch
Send a short description of the data, the crop or species and the decision you face. Within two working days, I reply with what your data can answer and which engagement fits, before you commit to anything. Your data stays confidential; an NDA is available on request.