Derek van Tilborg Molecular Machine Learning Scientist

Chemonaut
Derek van Tilborg profile picture

Hi! Iā€™m a PhD-trained scientist specializing in artificial intelligence for drug discovery, with both an experimental and computational background.

I have developed new deep learning methods and brought my models to the lab to discover new bioactive molecules, generate nanobiologics, and design nanoparticles.

You can find my research portfolio down below.

Out of distribution and molecular reconstruction

Molecular deep learning at the edge of chemical space
Derek van Tilborg, Luke Rossen, and Francesca Grisoni
Preprint (2025) šŸ”—
Code šŸ”—

I introduced a new method to estimate model reliability, based on molecular reconstruction. I experimentaly validated my method in the lab, resulting in the discovery of several new bioactive molecules for multiple target proteins.

Active learning for small molecule drug screening

Traversing chemical space with active deep learning for low-data drug discovery
Derek van Tilborg and Francesca Grisoni
Nature Computational Science (2024) šŸ”—
Code šŸ”—

I systematically benchmarked deep learning-based active learning strategies for low-data drug discovery, identifying key success factors and achieving up to six-fold improvement in hit discovery over traditional screening.

Active learning for nano particle design

Machine learning-guided high throughput nanoparticle design
Derek van Tilborg & Ana Ortiz-Perez et al.
Digital Discovery (2024) šŸ”—
Code šŸ”—

We developed an integrated platform combining microfluidics, high-content imaging, and active machine learning to optimize PLGA-PEG nanoparticles for cancer cell uptake, achieving a 3Ɨ improvement in just two iterative design cycles.

Activity cliffs

Exposing the Limitations of Molecular Machine Learning with Activity Cliffs
Derek van Tilborg, Alisa Alenicheva, and Francesca Grisoni
JCIM (2022) šŸ”—
Code šŸ”—

I developed an open-source benchmarking platform to evaluate machine learning models on activity cliffs, revealing that classical descriptor-based models often outperform deep learning methods in capturing complex structure-activity relationships. These benchmark datasets have been widely adopted in the field.

Nanobody design

AI-guided evolution of immune cell-specific nanobiologics
Manuscript in preperation

I designed nanobody binders that target a specific immune cell using protein language models. This project is in collaboration with Ayla Hokke, Koen de Bruin, and Senna Roelofs.