Your DNA has switches. I build the tools to read them.

Data scientist and PhD candidate in molecular and genomic epidemiology at Erasmus MC, Rotterdam.

Every cell carries the same DNA.

Around three billion letters, copied into almost every cell in your body.

So why isn't a neuron a liver cell?

Methyl groups, tiny chemical tags sitting on the DNA, help decide which genes stay active and which stay quiet.

The tags shift with life.

Pregnancy, smoking, growing up; methylation changes with what we live through, which makes it a record worth reading.

Reading that record is a data problem.

Hundreds of thousands of sites per sample, thousands of samples, and gaps everywhere. That's where I work.

See the work Get in touch

I like the unglamorous parts: missing values, batch effects, validation that quietly leaks. That's where results break without anyone noticing. Get those right, and the science gets to be interesting.

What I'm working on.

Research and software, built and validated on real data.

PhD research at Erasmus MC

Three conditions, one fingerprint?

Attention problems, BMI and lung function in children.

Generation R has followed thousands of Rotterdam children since before birth. Using their methylation data at birth and around age 9, I'm testing whether attention problems, BMI and lung function share epigenetic signals.

First paper in preparation.

Illustration of the idea, not study results.

Open-source tool

MethylImpute

Missing methylation values, filled in.

Real methylation data is never complete. Instead of dropping samples or sites with gaps, MethylImpute estimates the missing values so every measurement can stay in the analysis.

Check it out here
Illustration: as you scroll, the empty cells in a sample-by-site grid get filled in.

Randomised trial · Burkina Faso

What a mother eats, what her child carries.

A peanut-based supplement in pregnancy.

In a randomised trial in Burkina Faso, pregnant women received either a peanut-based nutritional supplement or iron-folic acid as control. I analysed what it meant for their children: their growth, and the epigenetic marks on their DNA.

Ghent University, World Health Organization, 2024

Illustration of the trial design, not results.

Python package

From config file to validated prediction.

A pipeline for building methylation-based prediction scores. One config file drives everything, and it runs on a compute cluster without babysitting. First use: estimating gestational age from cord blood.

  1. ConfigOne file describes the whole run.
  2. PrepareClean, filter and align the samples.
  3. Cross-validateNested, so tuning never sees the test data.
  4. ReportMetrics and plots for every fold.
  5. ApplyScore new samples with the final model.

The story so far

From a first dataset to a PhD in Rotterdam.

2024

Predicting CRISPR repair

A machine-learning pipeline that predicts how DNA repairs itself after a CRISPR-Cas9 cut (insertions, deletion length, indel diversity, frameshifts) from the 23-letter guide sequence alone.

Python · LightGBM ensembles

CRISPROutcome-ML on GitHub

2024

DeepMaize

Spotting maize leaf diseases from photos with a Swin Transformer: healthy leaves, common rust, blight and gray leaf spot.

PyTorch · computer vision

DeepMaize on GitHub

2025

MISAME III trial

Analysed a randomised trial of a peanut-based supplement in pregnancy and its effects on the children. A Gates Foundation funded project in collaboration with WHO-IARC and Ghent University.

2025

Built MethylImpute

A tool for filling in missing methylation data, so studies can keep every sample.

2025

Erasmus MC

Started a PhD in molecular and genomic epidemiology on the Generation R cohort, as part of the STAGE consortium.

What's next?

A good collaborations and a refreshing cup coffee to start with!

Have a data problem worth solving?

I'm open to collaborations, side projects and good conversations.

Email me LinkedIn GitHub