Kaspar Märtens

Hi! I'm a machine learning researcher developing foundation models for biology, with a PhD in Statistical Machine Learning from Oxford and expertise in probabilistic and generative modelling. At Synthesize Bio, I work on the models behind our virtual lab: systems that predict the outcomes of biological experiments, from cellular perturbations to human-level readouts. My recent work explores how LLMs can advance virtual cell models.

Scholar GitHub Twitter Email
Kaspar Märtens
Selected Work
LangPert: LLM-Driven Contextual Synthesis for Perturbation Prediction

LangPert: LLM-Driven Contextual Synthesis for Perturbation Prediction

MLGenX at ICLR 2025 · Oral
LLMvirtual cellreasoning
Enhancing Generative Perturbation Models with LLM-Informed Gene Embeddings

Enhancing Generative Perturbation Models with LLM-Informed Gene Embeddings

MLGenX at ICLR 2024
LLMvirtual cellGaussian Process
2026 GenePT Revisited: Do Better Text Embeddings Make Better Gene Embeddings? bioRxiv LLMembeddingvirtual cell 2025 SynthPert: Enhancing LLM Biological Reasoning via Synthetic Reasoning Traces Arxiv LLMpost-trainingreasoningvirtual cell 2020 BasisVAE: Translation-Invariant Feature-Level Clustering with VAEs AISTATS VAEclusteringhierarchical Bayes
All publications →
Writing
Neural Processes as distributions over functions
Aug 2018

Neural Processes as distributions over functions

What Neural Processes are and how they behave as a prior over functions — interactive visualisations included.