“The digital transformation can start small, too”

Radically accelerating the development of new materials with innovative technologies – that is the goal of the US start-up Kebotix. The Harvard University spin-off has recently become part of the 5-HT network. In an interview, Chief Product Officer and co-founder Dr. Christoph Kreisbeck, explains how Kebotix wants to help chemical and pharmaceutical companies save time and money in the development of new materials and at the same time take a step towards digital transformation.

Kebotix Christoph Kreisbeck

Which problems in material development does Kebotix address?

Many social and technological challenges in today’s world can be solved by new materials, for example by developing materials for energy storage, environmentally friendly plastics or novel drugs. We need solutions like these not just tomorrow, but today. But so far it takes ten to fifteen years to develop a new material. This problem affects various industries, including the pharmaceutical and chemical industries. For several years now, innovation productivity has been declining. Although we are spending more and more money on research institutions, output is getting smaller because the issues are becoming more complex and the development of new materials is therefore taking longer and longer. We want to make it possible to bring material products to market ten times faster than before. Our motto is: “Materials for tomorrow, today!

How does Kebotix accelerate the process of material development?

We pursue the concept of Inverse Design. When we want to develop a new chemical, we do not start with the structure, but with the function, which is then translated into a corresponding structure. At the moment, it is usually the case that the expert in the company has an idea, based on his experience, which material could have the desired properties. We go backwards – from the desired properties to the material. In doing so, we use new technologies for the creation of structures. An important starting point for speeding up the process is a kind of filter system: Along the pipeline through which the materials have to pass, we sort out more and more candidates from one step to the next, so that in the end only those candidates remain who have a relatively high probability of meeting all the criteria. This means that only a few promising materials need to be tested, which reduces costs and increases the success rate.

Kebotix_Self_Driving_Lab

What technologies are behind this approach?

The special feature of Kebotix is that we bring together different technologies in one integrated platform. On the one hand, we use big-data frameworks like those used by Facebook or Google for image and text generation. With Google Translate, for example, you enter a German text which is then translated into English. The basis for this is that the algorithm has been trained with large amounts of data to understand how the two languages are structured and how one can be translated into the other. We transfer this way of working to the translation of chemical properties into chemical structures: On the one hand, we enter the properties that the material we are looking for has to have, and on the other hand, a text comes out that describes the corresponding molecular structure according to a defined syntax. To do this, the algorithm must be trained with corresponding sample data to understand how chemical properties can be translated into chemical structures. To generate the necessary amount of data, we use a variety of tricks. For example, we integrate computer-aided simulation and machine learning to generate enough training examples for our algorithms from small experimental data sets. So we generate Big Data from Small Data. In addition, we also integrate findings from experiments that can be automated in the laboratory. Similar to us humans, our artificial intelligence is constantly learning from mistakes and successes. We believe that material development can only be accelerated if all these building blocks are integrated into one platform.

How can pharmaceutical and chemical companies in particular benefit from your services?

In the pharmaceutical industry, for example, we can help find suitable molecules that are able to bind to certain proteins in the early stages of drug development. In the agrochemical industry, for example, there is a great need for innovation in the development of fungicides, herbicides and pesticides. Here we can help to find new agents that are as non-toxic as possible and to which there is no resistance as yet. We also focus on optoelectronics, composite materials, plastics and lubricants. In all these cases, artificial intelligence can help to achieve the desired results faster. In one of our internal material development programs, we are working on suitable coatings for so-called smart windows, which contribute to energy efficiency in the building sector. It’s an exciting technology, but one that is not yet mature enough to survive on the market – so it’s an ideal problem to demonstrate what our technology is capable of.

How did the foundation of Kebotix come about?

Kebotix is a spin-off of a research group at Harvard University that is working on how the development of materials can be accelerated using digital innovations. In the research group of Prof. Alan Aspuru-Guzik, I initially worked with my co-founders Semion Saikin and Dennis Sheberla with computer-aided simulations. Soon after, we also integrated machine learning approaches. Because the algorithm was now able to learn the correlations between structure and properties, we no longer had to calculate everything, but could make better predictions with less effort. When we discovered that we could also use image and text generation technologies to generate molecules, we were ready for commercialization. That’s why we decided to found Kebotix together with Jill Becker, who had already set up her first own company. We started to scale very early and recently we completed our Series A round with $11.4 million. Kebotix now has 17 employees, from data scientists and machine learning experts to organic chemists and materials scientists.

Kebotix Team

What are the next plans for the further development of Kebotix?

We are currently growing across the board and addressing different industry segments to see where we can help with the digital transformation. In the near future, we want above all to acquire additional customers whom we can accompany on this path. It is important for us to communicate to our clients that we can start small: Companies often think that they first have to cleanly prepare all their data before they can start with a machine learning project, but this is not true. We don’t have to start with a big revolution – digital transformation can also be a gradual evolution. Our customers don’t have to do everything from scratch, but if our tools help them to increase productivity in certain areas, that’s already a significant gain.

How strong is your focus on Germany in the search for new customers?

Many of our current customers come from the USA, Japan or Germany, where many companies are already quite open-minded about artificial intelligence. Germany is also a particularly important market for us in the future. The German chemical industry in particular is very strong and is known worldwide for its innovative strength. We are currently already talking to several large chemical companies in Germany about potential projects. By working with 5-HT, we therefore hope to raise awareness of the potential of materials informatics and artificial intelligence for materials science in the German market and gain better insights into how we can best support local companies in their digital transformation. The Rhine-Neckar region with Ludwigshafen as the hotspot of the chemical industry is a very important region for us. Here we see many opportunities to exploit synergies and create added value for the ecosystem.

Be part of the 5-HT Digital Hub Chemistry & Health

Get talking to innovative startup and established corporates in our network.