Dr. Lars Gabriel

I am interested in research that moves from meaningful biological questions into methods people can actually use. Much of my work focuses on turning complex biological evidence into practical annotation tools that support real genome projects.

In recent years, I have worked primarily on eukaryotic gene structure prediction, deep learning, and the broader open-source annotation ecosystem around Tiberius, BRAKER3, and TSEBRA.

I enjoy work that sits between method development and usable software: models should be scientifically well-motivated, but they also need to run reliably, scale to real datasets, and produce results that biologists can inspect and use.

My academic background combines biomathematics, mathematics, bioinformatics, and computational biology. I studied biomathematics in my bachelor's program and continued with a master's degree in mathematics before focusing my doctoral and postdoctoral work on eukaryotic genome annotation, especially machine learning methods for automated gene prediction and the integration of RNA-seq, protein, and genomic sequence evidence.

Working daily on high-performance computing systems has also shaped how I think about research software. I care about tools that are accurate, reproducible, documented well enough for real users, and robust when applied beyond the dataset they were developed on.

My current research centers on deep learning for gene prediction across diverse eukaryotic clades. I am interested in how sequence-based models can learn gene structure, how biological rules can inform deep learning methods for biological tasks, and how annotation quality can be evaluated across different organisms.

More broadly, I am interested in applying machine learning to biological sequence analysis tasks where the result can become a scalable and useful scientific tool, not only a model benchmark.