ProtAl: An active learning platform for protein engineering

Name of participant: Sebastian Franz

Project’s name: ProtAl: An active learning platform for protein engineering

Project description:

Proteins are fundamental components of organic life and play a pivotal role in biological processes. They are used increasingly for medical and biotechnological purposes, ranging from drug development and agricultural technologies to precision fermentation.
Proteins often need to be adjusted to specific conditions, to achieve optimal results. Improving proteins (protein engineering) acts as one of the biggest challenges for modern biotechnology and bioinformatics: With small data sets, desirable changes (mutations) are supposed to be discovered in an efficient and cost efficient manner.

Our Software Campus project ProtAl connects active learning with the up-to-date methods of bioinformatics based on protein language models (such as ProtT5 or ESM-2). Active learning is an iterative process, during which new mutations are suggested and tested in the laboratory. This way, the model learns as efficiently as possible which mutations work best. The protein language models, in turn, offer credible representations of proteins, that are crucial for the successful research of mutations. This enables them to generate particularly promising mutations, instead of leaving that to pure chance.

Further, our project aims to make the methods we developed broadly and easily accessible. Instead of simply offering black-box methods, as is often the case in artificial intelligence research, we want to present all stages of the active learning process in a way that is transparent and understandable to researchers and users.
Through this, the results are easier to interpret, and the methods are more adjustable to specific conditions. This also makes it possible to draw on specific human expertise and experience to manage the process effectively. We achieve this accessibility through a modern user interface that is integrated into the biocentral project. The resource-intensive calculations are carried out on our cloud service to further improve the user experience for researchers.

Together with our collaborative partners, the Merck KGaA, we further plan an intensive benchmarking, to assess the benefits of our ProtAl framework and continuously improve it. Finally, the project will be topped off by a laboratory test in which we aim to optimise the activity of a specific protein (enzyme) using our methods. We are delighted to have secured the Jimenez-Soto research group at LMU Munich as a collaboration partner for this.

Software Campus Partner: TU München and Merck KgaA

Implementation period: 01.05.2026 – 31.10.2027