From bioethanol to laundry detergents, the biotech industry is developing enzymes that make everyday processes greener. During enzyme development, we engineer specific mutations to obtain desired properties, nowadays, often with the use of machine learning (ML) methods. However, it has become increasingly difficult to track which models perform best for our property of interest. Together with IFF, the Marks lab and Xebia, we developed a community driven framework that makes benchmarking as simple as a pull request. We present two packages to facilitate that process, proteingym-base and proteingym-benchmark. In proteingym-base we standardize the collection of biochemical data modalities into a single data-science ready archive, while proteingym-benchmark provides the setup for automated model benchmarking.

From bioethanol to laundry detergents, the biotech industry is developing enzymes that make everyday processes greener. During enzyme development, we engineer specific mutations to obtain desired properties, nowadays, often with the use of machine learning (ML) methods. However, it has become increasingly difficult to track which models perform best for our property of interest. Together with IFF, the Marks lab and Xebia, we developed a community driven framework that makes benchmarking as simple as a pull request. We present two packages to facilitate that process, proteingym-base and proteingym-benchmark. In proteingym-base we standardize the collection of biochemical data modalities into a single data-science ready archive, while proteingym-benchmark provides the setup for automated model benchmarking.