Artificial intelligence for Genetic Evolution Mimicking
To simplify the generation of desired functionalities de novo and enable evolutionary leaps, we created AiGEM, our Artificial Intelligence for Genetic Evolution Mimicking software suite. The heart of our software is DeeProtein, a deep neural network trained on ~10 million protein sequences and able to infer sequence-function relationships from raw sequence data with high accuracy. By interfacing DeeProtein with our genetic algorithm GAIA, we – for the first time - established a fully closed, in silico evolution cycle driven by an intelligent network. We used AiGEM to successfully modulate the catalytic efficiency of β-lactamases, thereby even producing variants which highly outperform the wild type enzyme. To demonstrate AiGEM’s ability for generating functionality de novo, we created a novel, highly efficient β-galactosidase purely by in silico evolution of a β-glucuronidase parental sequence. Finally, as part of our integrated human practices project, we implemented SafetyNet, a DeeProtein-based web application safeguarding directed evolution experiments.
Card image cap


We apply deep learning models to harness the complex sequence to function relation in proteins.

Card image cap


By interfacing our trained models with a genetic algorithm we developed an in silico evolution tool.

Card image cap


To perform self-checks, and prevent misuse of directed evolution techniques, we developed SafetyNet a sensitive tool for the detection of harmful traits in sequences.

Card image cap


Deploying GAIA, we fully in silico evolved a beta lactamase and reprogrammed the E. Coli beta glucuronidase towards beta galactosidase function.

Card image cap

MAWS 2.0

The availability of specific riboswitches is key to the succes of enzyme PREDCEL/PACE experiments. We therefore reiterated the MAWS aptamer prediction software.