INTERVIEW
Case Study Interview
Research and development that would have taken two years shortened to just a few weeks: AI discovers conditions for high production of structural proteins in a short period of time
Background
Although the company had been studying ways to increase productivity, the number of combinations of experimental conditions was enormous, and research and development took years.
As a venture company, we had to invest limited funds in research and development and needed to produce results in a short period of time.
Effects after implementation
Research and development results that previously took two years can now be achieved in just a few weeks.
We were able to discover a combination that no one had ever thought of. Therefore, by getting to the bottom of why it works, new discoveries may be made.
It has become possible to visualize what AI can do, and the scope of what can be done (by combining existing research and development with AI) has expanded.
Spiber Inc. is a bio venture that develops and produces structural protein "Brewed Protein™ material." Brewed Protein™ fiber is environmentally degradable and is expected to be a next-generation material that can provide an alternative solution to conventional animal-derived, plant-derived, and synthetic materials. 71 products are sold by 16 brands, including The North Face and Sacai. We spoke with Toru Takahashi, head of the biotechnology division, about the background to the introduction of Epistra Accelerate and the changes that have occurred since then.
As I was cashing out moment by moment, I wanted to thoroughly research ways to achieve results in a short amount of time.
- First of all, please tell us about the role of the department that you oversee.
The Biotechnology Division is a department that thoroughly develops technology to produce the protein raw materials that we at Spiber develop and manufacture at low cost and with high quality. Specifically, we conduct research and development of all processes for producing protein raw materials using microorganisms, from protein sequence design, strain breeding, fermentation production process, purification process, scale-up, and actual machine prototyping. Currently, Spiber has about 300 employees, of which 40 members are registered, making it a relatively large division.
- It's a department that acts as the engine of a cutting-edge biotechnology company. What challenges did you face before introducing Epistra Accelerate?
As I mentioned earlier, our division's theme is how to produce high-quality protein materials at low cost. To achieve this, we use our past knowledge and human-wave tactics to test and examine various conditions. However, to determine the culture conditions for one protein, we need to test the conditions approximately 200 to 300 times, which takes years.
Moreover, new proteins need to be produced one after another depending on the purpose of use, so we couldn't spend a long time on just one protein.
Furthermore, as a venture company, we are conducting research and development with limited investment funds, so every day is precious. I have always wanted to somehow reduce the number of experiments and thoroughly research ways to produce results in a short period of time.
.jpg)
- Why do we need to consider so many conditions in the first place?
Let me use the study of culture conditions (the first case study of Epistra Accelerate) as an example. Basically, the field of microbiology has been around for a long time, so we know to some extent what conditions microorganisms need to grow. For example, pH should be within this range, and temperature should be within this range.
So what we do is consider parameters such as pH one by one based on the conditions that are already known.
However, there are a huge number of parameters, and even if you combine the best conditions considered one by one, the best results may not be achieved, and in fact the performance may decrease. For this reason, it is unavoidable to conduct the study many times.
As living organisms, they will continue to live in the way that is easiest for themself, even if other external factors change, so creating things through cultivation doesn't work as well as humans think.
Right: Toru Takahashi, Head of Biotechnology Division, Spiber Inc. Left: Yosuke Ozawa, Ph.D., Co-founder and CEO, Epistra Inc.
I was skeptical about the effectiveness of introducing AI into bioprocess development, but I wanted to give it a try.
- Facing such an urgent issue, did you search for a solution?
Yes. At the time (around 2018-2019), I often heard the words AI and machine learning, and information about how amazing they were was constantly coming in. Even though I was in the bio field, I heard and read in various places that it was effective to have AI learn big data and derive optimal solutions.
However, we were still skeptical about whether the technology could be applied to the issues we faced.Also, we were in the midst of various attempts to solve the issues, such as joint research with academia and large companies, so we were not thinking too much about using AI.
-What made you decide to work with Epistra?
It all started when I happened to meet Mr. Ozawa, the CEO of Epistola, and we talked about whether he would be interested in optimizing culture conditions using AI. When I asked him about it in more detail, I learned that in a joint research project with the RIKEN Institute, they had been somewhat successful in optimizing differentiation induction conditions on the theme of iPS cell differentiation.
Although I was skeptical, I decided to try it out for a set period of time, because it had already produced some results in the field of biotechnology and there was a possibility that it could lead to "improving protein productivity in a short period of time," which was a very important issue for Spiber.
I thought the era of humans would come to an end because AI has come up with conditions that humans would never think of.
-What are your impressions after introducing Epistra Accelerate?
I was surprised that the results came much sooner than I had expected.
After training the AI to learn from past data and culturing the cells 10 to 20 times, the amount of protein produced jumped up significantly. When I first heard the report, I even suspected that the quantitative assessment of the protein amount was incorrect, and instructed them to do it again.
We were also surprised at the highly productive culture conditions that the AI came up with. Humans tend to set values that are easy to round off, but the AI set values much more precisely, and the accumulation of these small differences ultimately produced big differences. I'm sure all the lab members thought, "No human could ever come up with something like this."
Witnessing how high productivity could be achieved in a short period of time under conditions that humans would never have thought of made me realize that "the era of humans in certain fields is coming to an end." It may be fine for humans to create the base conditions (the initial conditions that work as the starting point for exploration) because of past experience. However, I realized that it would be better to leave the optimization beyond that to AI.
As a researcher, I do feel a bit sad, but as the person in charge of R&D at a company, I believe it is more important to contribute to society by selecting and actively introducing reliable technologies and accelerating the progress of our business.
Thanks to the existence of a common language and high-quality support, the project produced results from the early stages
- I'm glad you were pleasantly surprised. On the other hand, were there any moments that confused you?
There was nothing in particular that happened. I think it was actually Epistola who was confused.
The first important step in utilizing AI is to train it on past data. However, although our data was in a format that humans could understand, it was not in a nice format that could be used by AI to learn from. In addition, during the relatively long research and development period, there were also minor issues such as changes in the items being recorded and missing values.
However, Epistola helped us solve the problem by creating a machine-readable format, confirming the meaning of the data, and figuring out appropriate ways to fill in missing values. Thanks to her help, we were able to solve the problem smoothly from the start.
They anticipated the areas where we might get stuck and helped us out, so we didn't get confused as much as we should have.
-What do you think was the reason that this cross-disciplinary effort between data science and biology was successful from the start?
One reason is that there was a certain degree of common language from the beginning.
When working together in different fields, you need a common language. Without it, you have to start by learning each field, which can waste about six months of time. This can make it seem like the company has invested a year but nothing has progressed, which can make it difficult to continue the project.
However, Mr. Epistola has a solid knowledge of not only data science but also biology. Being able to smoothly communicate basic concepts is something that is often overlooked, but I think it is an extremely important element.
Another reason is that we were able to trust them and receive their generous support. Even if we share a common language, when we actually begin to work on a project, it is inevitable that we will come across a misunderstanding.
Epistola has a very keen sense of how to spot situations where we don't understand each other. When it seemed like the other person didn't understand, he would be creative with the wording and work closely with the other person to move the project forward, so even if there were misunderstandings, they didn't end up becoming big problems.
Also, if you share your objectives with them beforehand, they will use their expertise in other fields to come up with new proposals when there are obstacles to moving the project forward.
.jpg)
-Finally, please tell us about your future prospects for research and development using AI.
By using AI once, we were able to get a good idea of what cutting-edge AI technology can and cannot do. As a result, we can now concretely imagine how combining AI with problems other than those we first applied to could be solved. This has greatly broadened the scope of what we can do in research and development.
Spiber's strength is that it has a comprehensive process from protein design, cultivation, and purification to fiberization and post-fiber processing, and it has accumulated technology and data at each layer.
By combining this with Epistola's AI, we would like to further accelerate our research and development. If the technological development we are currently planning bears fruit, I believe it will revolutionize bioprocess development, so I am looking forward to it.
Recommend
Other recommended articles
Research and development that would have taken two years shortened to just a few weeks: AI discovers conditions for high production of structural proteins in a short period of time
Spiber Inc. is a bio venture that develops and produces structural protein "Brewed Protein™ material." Brewed Protein™ fiber is environmentally degradable and is expected to be a next-generation material that can provide an alternative solution to conventional animal-derived, plant-derived, and synthetic materials. 71 products are sold by 16 brands, including The North Face and Sacai. We spoke with Toru Takahashi, head of the biotechnology division, about the background to the introduction of Epistra Accelerate and the changes that have occurred since then.
Read the article
Searching for cell recipes for regenerative medicine using AI. Achieving efficiency equal to or greater than that of an expert.
Masayo Takahashi was the first in the world to successfully perform a retinal transplant using iPS cells in September 2014. She is currently the CEO of Vision Care Co., Ltd. and continues to work to bring cutting-edge treatments to many patients. We spoke to her about the background to her collaborative research using Epistra Accelerate and her future prospects.
Read the article