Train predictive models

We implement predictive models that can be used to predict target values based of quantitative or qualitative inputs. These predictive models can be used to predict the phenotype of different biological products, such as the binding affinity of an antibody or the titer of a metabolite. Biological and chemical entities can be highly dimensional but can be mapped to an embedding space. We implement different embeddings to represent DNA molecules, proteins, and chemical compounds, to efficiently train predictive models.

Generate new leads

Our generative  models can be trained to learn the probability distribution of the  molecules in a training dataset, and then used to sample it and generate new leads that follow the same distribution.

We have shown that by using some generative models we can capture the desired properties of certain peptides and protein domains.

“In collaboration with TeselaGen, at DTU we have used machine learning models to generate new design recommendations, enabling us to successfully forward engineer the aromatic amino acid metabolism in yeast.”
Michael Jensen
Michael Krogh Jensen, PhD
The Novo Nordisk Foundation Center for Biosustainability

Run evolutive models

We implement bayesian optimization techniques to recommend new experiments  based on empirical data gathered in the lab.

Our evolutive models can use other pre-trained predictive models as surrogate models. Evolutive models can also be used with pre-trained generative models, when exploring highly dimensional design spaces.

Deploy your own AI models

Our platform implements advanced deep learning  techniques  to model biological entities, and to optimize synthetic constructs of DNA, proteins, and other biomolecules.

Train our own off-the-shelf models, or integrate our operating system with your own  algorithms.

 

Are you an academic group? A startup? An enterprise?

Don’t worry, we have something for everyone. Schedule a demo today.