AI-driven Drug Discovery is already signigicantly high level

AI is used clinical trials including Drugs Pipeline discovery & verification, the key point is Data

Prof. Kang, Jaewoo, Dept. of Computer Science and Engineering, Korea University, introduced the case at 'BIOplus 2019'

Global pharmaceutical companies are showing great interest in the development of new drugs using artificial intelligence (AI).

 

It is expected that AI will drastically reduce the time and costs associated with finding a candidate substance through clinical trials and approvals before entering the market.

 

What about the perspective of the computer science industry, not the pharmaceutical industry?

 

Prof.

 

Kang, Jaewoo, Dept.

 

of Computer Science and Engineering, Korea University, explained that AI-driven Drug Discovery is already coming to reality at 'BIOplus 2019' held at COEX in Seoul on Nov 12.

 

Prof.

 

Kang presented as a speaker at a session called ''AI Medicine: Data-driven Drug Discovery.

 

He expects AI to be particularly strong in clinical trials including Drugs Pipeline discovery & verification.

 

In addition, he introduced cases of this and added his experience of winning the global competition continuesly, According to him, the most prominent step in AI development is Drugs Pipeline discovery.

 

At present, the discovery of candidate materials is mainly done by manual method through document retrieval like Pub Med, the world's largest medical thesis site, reads related articles and formulates hypotheses.

 

However, the number of new articles published in Pub Med is 5000 conservatives a day.

 

It is impossible for a person to check one by one.

 

AI can contribute to this, Professor Kang explained.

 

He focused on AI's "human language learning ability." He succeeded in creating a program called 'BioBERT' last January based on Google's 'deep learning' technology.

 

AI ​​will answer it.

 

if you enter the medical question you want in the program.

 

It is an expansion of the pharmaceutical bio sector of 'BERT' which Google disclosed earlier.

 

Google learned Wikipedia's language with deep learning and created a program called BERT.

 

Data was entered as Wikipedia's language.

 

Instead of Wikipedia, BioBERT learned the language of Pub Med and PMC (another medical article search site).

 

16 billion words were entered as data.

 

BioBERT's capabilities have been recognized globally.

 

He won first place in BioASQ, one of the AI ​​bio competitions in the pharmaceutical bio sector.

 

Google was in second place.

 

It came with Google's model and reborn as a better program than Google.

 

He said,“It was a test to see how accurate the program would answer, and we can significantly reduce time and effort to find candidates (via BioBERT)." Next step is to verify it once you formulate a hypothesis.

 

Prof.

 

Kang also explained that AI contributes a lot.

 

He added real experience.

 

It was a competition called 'IDG-DREEM' this year.

 

When professors at Mount Sinai Medical School in New York presented their hypothesis as a problem, AI experts around the world searched for answers.

 

Professor Kang won the championship for two consecutive years.

 

He became a co-winner this year.

 

It is in the same position as Illinois State University, China National University of China, and North Carolina State University.

 

The problem was that 'find ZINC15 (Pharmaceutical Bio-Sector Compound DB)' to determine what is most effective when combined with certain drugs in thyroid cancer.

 

In the case of drugs with different structures and similar mechanisms, Prof.

 

Kang's group trained machine learning models to recognize as similar drugs.

 

As a result, It was successful to find 10 substances.

 

Seven of these substances were also confirmed to have related papers.

 

AI could also help with clinical trials, Pro.

 

Kang predicted.

 

"We can predict which patients will be particularly effective for which patients or for which biomarkers," he said.

 

In 2016, AstraZeneca held a contest with Sanger.

 

After revealing the patient's genetic information and the effects of the drug, the problem was predicting which cancer would be most effective.

 

Prof.

 

Kang said that more data is required than the previous stage of finding and verifying candidates.

 

Even though it was the latest professional computer at that time, it took over 20 hours to solve a problem on its own.

 

Eventually, 20 more computers were purchased and the processing speed was increased by 20 times and then reduced to 1 hour.

 

Prof.

 

Kang's group, who first entered the competition, placed second .

 

The following year, he was ranked No.

 

1 in a competition hosted by the National Cancer Institute (NCI) in 2017.

 

"In 2016 and 2017, we were able to succeed because we used the good quality of patient information that was organized in advance by the organizer." "It is very important to reine data for AI and Even in the field, quality & quantity of information is very important.“

"AI also learns from data… the more the better" In the subsequent presentation, the importance of the data was emphasized.

 

Ph.D Shin, Hyun-jin, who is in charge of AI drug development at Takeda Pharmaceuticals, said, "I feel like I'm floating in the ocean.

 

It's water everywhere but there's no water to drink." There is a lot of data, but no useful data available.

 

"In order for machine learning to evolve, standardized data must be available.

 

However, different data formats are difficult to use these days.“ “Dissemination is the most important to win the war in the long run”he said.

 

"It is same as AI drug development.

 

Both the quantity and quality of the data is important.

 

lower quality may result in lower accuracy, while lower quantities may result in biased results“ He emphasized.