[Basic Information] What is AI Drug Discovery?
November 20, 2023
New Drug Development Status & News from Pharmaceutical Companies Manufacturers
November 20, 2023
【Interview cooperation】What FRONTEO's AI
AI?
This page introduces FRONTEO, Inc. which provides drug discovery support AI services leading more efficient, faster, and cost-compressed basic research, target discovery, hypothesis generation, and drug repositioning in new drug development.
What Different FRONTEO's AI
1. What Different FRONTEO's AI
What FRONTEO
1.1 What FRONTEO
FRONTEO was founded in August 2003, and its main business is data analysis utilizing artificial intelligence. In addition to our original business of legal tech, the company has developed businesses such as life science AI, business intelligence, and economic security. Our main customers are government agencies, law firms, medical institutions as well as private companies, and we are also a global company with offices in the United States, South Korea, Taiwan, and other countries.
In the legal tech AI business, the company has more than 10,600 cases (as of June 2023), including U.S. Department of Justice investigations, U.S. civil lawsuits, and domestic fraud investigations. In the Business Intelligence business, we boast a cumulative total of 328 companies (as of June 2023) in industries such as financial institutions, manufacturing and construction, and life sciences.
In the life science AI business, our proprietary natural language AI is utilized to support significantly streamline, accelerate, and improve the probability of success in drug discovery research through the fusion of AI and drug discovery experts.
KIBIT, a Dictionary-Free Natural Language AI
1.2 KIBIT, a Dictionary-Free Natural Language AI
KIBIT is an AI engine developed by FRONTEO. It is capable of effectively analyzing and utilizing medical data, including a large amount of text data with free descriptions, based on evidence (rationale). Its strengths and features include the following :
Dictionary-free
No language restriction
Co-analysis of text and numerical values
Parses words and sentences together and realizes them in approximate formulas
With these strengths, KIBIT is classified as a distributed representation type of natural language AI, and is capable of making judgments based on complex combinations, which has been difficult with conventional dictionary/scenario type systems. It can also cover data that changes over time, and can cover a wide variety of terms without being limited by keywords.
Overcoming the Challenges of Natural Language AI
1.3 Overcoming the Challenges of Natural Language AI
AI in drug discovery research has been used in processes such as seed generation and optimization,pre-clinical trials, and clinical trials, but there has been little AI that can be used in the initial stages of idea conception, hypothesis generation, and drug target identification and analysis. In the past, there was a movement to utilize natural language AI for target discovery, but it was difficult to overcome the following issues.
Requiring enormous computer power
Difficult to manage and maintain the thesaurus (dictionary)
Even biologists cannot make judgment with results
FROTEO's drug discovery support AI service successfully overcame the above challenges. The system is light enough to be used on a laptop and does not require a thesaurus (dictionary). In addition, biologists with both experience in pharmaceutical companies and knowledge of AI can make proposals tailored to each customer's order. This system of drug discovery supported by AI and biologists is collectively called the Drug Discovery AI Factory.
What is the Drug Discovery AI Factory?
2. What is the Drug Discovery AI Factory?
FRONTEO's Drug Discovery AI Factory does more than just use AI to solve drug discovery challenges such as basic research, hypothesis generation, and target discovery. Biologists who have experience in drug discovery research at pharmaceutical companies and research institutes and are well versed in AI will utilize KIBIT and various AI applications to propose target molecules, biomarkers, MoAs, and new indications in a short time and at high throughput using five unique analysis methods: common-unique pathway analysis, two-dimensional mapping analysis, Vector additive analysis, multifaceted analysis, and virtual experiments.
[Special Interview] Interview with the person in charge
What can only be achieved with FRONTEO’s AI/services
Hiroyoshi Toyoshiba, Ph.D., Chief Technology Officer, Corporate Officer, will explain in detail the appeal of KIBIT, a natural language AI.
Hiroyoshi Toyoshiba, Ph.D., Chief Technology Officer, Corporate Officer, FRONTEO
Q. Please tell us about what natural language AI like KIBIT can do and its strengths in drug discovery.
A. Covering both artificial intelligence and biology
[Toyoshiba]: There are other companies besides ours that are utilizing linguistic AI for drug discovery, but simply using linguistic AI does not mean that drug discovery will be successful. To realize even hypothesis generation, which is an important process in drug discovery, knowledge, and resources in both artificial intelligence (information systems) and biology are necessary. FRONTEO's strength lies in our ability to cover both.
In particular, biology has a history that has been passed down from generation to generation. It is difficult to connect this historical academic field with the new perspective brought forth by artificial intelligence (information systems). The reality is that there are few teams or companies that can produce output that satisfies biologists. Our company is unique in the sense that we are able to produce output that balances these two elements at a high level.
Q. I understand that KIBIT is classified as a distributed representation type of natural language AI, but what specific advantages does it have?
A. Unexpected outputs are produced.
[Toyoshiba]: In distributed representation type, it is following the rule that the meaning of a word is determined by what words appear around it and how often. Compared to the dictionary/scenario type, the distributed representation type, is characterized by the fact that it produces output that never expected. This is important. We tend to stick to what we have done and seen so far. This is bias.
The greatest advantage of the distributed representation type is that, as mentioned above, it is based on statistical information and thus produces output from a flat perspective. In other words, the point is to make us aware of cognitive bias. However, no matter how suggestive the output of AI may be, it is meaningless unless humans receive and utilize the results. I would like people to look at the results in a flat way and accept them as generally true, rather than saying, It's not good because it's different from what I feel.
Q. Why did you decide to develop such an AI in the first place?
A. It can always produce the same optimal solution.
[Toyoshiba]: There have been distributed representation type AI before, but we developed KIBIT on our own because we did not want to use an engine created by another company. We wanted to provide ideal solutions with AI developed based on our own theories.
We were also concerned that other companies' distributed representation AI would produce slightly different solutions each time. By incorporating a mathematical approach, KIBIT is tailored in such a way that the same optimal solution is always output.
