 In our latest work, we trained a machine learning model that can support scientists to pick the right enzyme that catalyze a chemical reaction. This is fundamental because using enzymes as catalysts can help lower the barrier of adoption of sustainable processes in industrial chemistry. Enzymes are molecular machines that nature optimise in the last 3.5 billion of years to speed up chemical reactions in our body and in any laboring organism. The main advantage of using enzymes is that they usually operate under mild conditions, in water or at body temperature. Furthermore, just like the proteins in us or in our food, they are made up of amino acids, which makes them essentially compostable. Compared to these traditional approaches to speed up chemical reactions in industrial applications, often involves the usage of toxic solvents, high pressures, high temperatures, as well as a large amount of waste. Our data-driven model can help to bridge this gap by distilling the knowledge of biochemical reactions and help to harness these molecular machines that have been optimised through billions of years of evolution. In our project, the first challenge we have to face is that there is no single public source for biochemical data. To overcome this problem, we parse a wide variety of available data sources and we compile a public open data set of standard biochemical reactions. We call this ECREACT and we make this publicly available together with our paper. While ECREACT covers a wide variety of enzymatic reactions, it was still not enough to feed our model. To design, we use a technique called multi-taster learning to fine-tune our organic chemistry models to teach them enzyme chemistry. Our model is available on the RxN4 chemistry platform to ease the usage by the research community. Let's check an example together. It's mostly interesting to compare an organic chemistry model with an enzymatic chemistry model. So, when we are on the platform, we can simply upload the molecule that we want to synthesise and pick the model that is using enzymatic chemistry. Now, we will check an example of an expansion in split-screen where we see on the left the organic chemistry model and on the right the enzymatic chemistry model. We can immediately notice how the enzymatic chemistry model is suggesting routes that are indeed recommending the usage of enzymes to catalyse the reaction. Interestingly, our model was even able to spot errors in the ground through helping us to better compile our data sets and this testifies how well the model was able to distill biochemical reaction knowledge. If you want to explore our model, try it directly on RxN4 chemistry or play with it using the Python client available on GitHub.