Hello everyone, My name is Kenshin Fujiwara and I am the CEO and founder of HACARUS Inc.
Through this series of blogs, I will discuss a wide range of topics of AI, from the history of AI to practical tips for the successful application of AI projects. I hope my blog posts will help you better understand AI and solve your business issues.
In the last blog, I introduced the case of Mitsubishi Tanabe Pharma and what challenges they faced in the application of deep learning technology for drug discovery. Today, I would like to explain how the challenges were solved by using Sparse Modeling.
After weighing the pros and cons of deep learning, Mitsubishi Tanabe Pharma Corporation has decided to create an AI model using a different approach. Using sparse modeling methods instead, Mitsubishi Tanabe Pharma was able to solve all three of the aforementioned issues as below:
- It takes a long time to obtain results
- The AI judgements are hidden behind a black box.
- It requires a huge number of prediction models to build each compound (requiring a large amount of data)
Sparse Modeling Opens a New Door for Drug Discovery
First, sparse modeling was able to drastically reduce the time required to complete the project. On average, a compound analysis could take 15 to 40 minutes to complete, however, by utilizing sparse modeling, this process can be performed in only 16 seconds.
This has resulted in a model that is over 56 times more efficient. The reason for this improvement is the model’s ability to function with small amounts of data. At the same time, the accuracy is equal to or better than deep learning methods. This example is one case that shows the superiority of sparse modeling in specific applications.
To address the issue of black boxing, I want to explain a small experiment that we conducted involving the imaging of cells. In this case, sparse structure learning was used for a control group of normal cells and cells that had been treated with the compound. Sparse structure learning allowed us to interpret the correlation of features in both images and their change after being influenced by the compound. This helped illuminate the mechanism of action (MOA) of the compound.
The third challenge, the mass construction of prediction models, has also been solved since sparse modeling can operate with only a single prediction model that can handle all compounds.
As a part of the project, HACARUS was contracted to help Mitsubishi Tanabe Pharma with its drug discovery screening. During the course of this project, we had numerous opportunities to communicate about the challenges of deep learning.
Challenges of the Pharmaceutical Industry
In the pharmaceutical industry, drug prices are set by the government, which has decided to update prices every year since 2021. This opens the potential for their products to decrease in price every year. However, drug discovery is a long-term business and testing can last up to a year. In such an increasingly challenging environment, pharmaceutical companies must constantly strive to develop new drugs. Now, they don’t have the luxury of relying on human-based research which takes time and money.
For Mitsubishi Tanabe Pharma, the resolution of the three previously mentioned issues was a welcomed outcome. However, more than that, the establishment of a successful AI-based screen process was a major milestone for the company and the pharmaceutical industry as a whole.
In the next blog, I will introduce another case of an electronic equipment manufacturer which faced challenges in reducing inspection times. For your updated knowledge and insight about AI technology, subscribe to our newsletter or visit HACARUS website https://hacarus.com.
