Still fighting algorithms and computing power?

At the beginning of the 21st century, artificial intelligence (AI) technology made important breakthroughs and began to be truly applied to many fields, changing the development trajectory of many industries

From early detection to downstream, there are almost AI figures

To understand the application of a technology in a certain industry, you will inevitably pay attention to the field and direction of the use of this technology

"From the perspective of the entire drug R&D chain, the current AI is most concentrated in the field of early discovery

Jingtai CEO Ma Jian

In an article on the keyword statistics of the English literature of AI medicine, the 7 AI medicine research and development areas have the highest frequency, namely, Drug Targets Development, Drug Mining, and Compound Screening.

Specific research directions of artificial intelligence in the field of pharmaceutical research and development

Lai Caida, the co-founder of Jitai Pharmaceuticals, obviously agrees with this view.

In addition to early discovery, AI applications are currently expanding to more links in the industry chain, and AI startups are carrying out corresponding work in almost every link

In Lai Caida's view, the expansion of AI in the pharmaceutical chain focuses on application scenarios, and whether there is big data generation is the basis of application scenarios

Lai Caida, Co-founder of Jitai Pharmaceutical

"Molecular discovery and synthesis have many directions that can be carried out.

Data "touchstone"

Build the ability to obtain real data and stronger experimental capabilities

When talking about AI, it is inevitable to discuss three elements: algorithms, computing power, and data

As can be seen from the above, the entry direction of AI in pharmaceutical applications is a more data-rich link, and its development depends on the generation of data

"Among the three foundations of AI, there is no problem with computing power at present, and the algorithm will be optimized in the process of continuous witnessing.

Ma Jian said that the application development of AI technology itself is undergoing phase changes from algorithms, computing power to data

Now that AI-driven drug R&D has entered a critical midfield stage, the focus of this stage is to build the ability to generate and obtain real data, and to build stronger experimental capabilities.

Whether it can meet the requirements of data volume and quality has also become a touchstone for judging whether the development of new AI drugs has value
.

In fact, data itself is a complex system involving many influencing factors.
Not only the amount of data, but also various factors will affect the final AI application.
This is also the challenge
.

Among them, the quantity and quality of data is a very important issue
.
Ren Feng said that at present, most of the public data that everyone uses are uneven.
Due to the different experimental methods used by the company, the data consistency of certain specific parameters may also be relatively poor
.
One direction of future development is to reduce human variables through the data needed for some automated experiments
.
In addition, as the financing scale of AI companies increases, they can collect some unique data according to their own needs and expand the amount of data to further improve his algorithm
.

Data propensity is a matter of great concern to Ma Jian
.
He said that the information obtained from experiments, calculations and simulations, and public information all constitute data sources, but AI development requires some more standardized data.
If these data samples from a variety of sources are inclined or poorly diversified, It is impossible to train the AI ​​to describe the whole problem completely
.

"For example, most of the data obtained in the literature belong to positive samples, which are the results of successful research releases, but in fact, trial and error in a large number of studies is very valuable data.
Both negative and positive samples exist, and the understanding of this problem is better.
Complete
.
” Ma Jian told developers that in addition to fewer negative samples, there is also the problem of false positives.
Some experiments are not reproducible.
Therefore, publicly available data sources must be filtered a lot
.

In addition, there is also the problem of data migration
.
Comparing Ma Jian with automobile manufacturing, if a company that is good at manufacturing trucks wants to manufacture sports cars, although they all manufacture cars, the requirements for appearance and aerodynamics are completely different.
These constitute two very different fields.
Then truck design How valuable the data obtained in the process will be for the manufacture of sports cars
.
In the pharmaceutical field, when developing different indications and different targets, the characteristics of the problem are very different, which means that it is not whoever has the most historical data can develop faster
.

Waiting for verification from the pipeline

Deeply penetrate the industry, waiting for a new tipping point

In the past two years, AI companies that have made rapid progress have announced their pre-clinical candidates designed through AI
.
In February 2020, the British company Exscientia announced the development of the preclinical candidate compound DSP-1181 for the treatment of obsessive-compulsive disorder.
In March of the same year, the project entered the clinical trial phase
.
In 2021, Anglo Silicon has announced two preclinical candidate compounds for fibrosis.
The fastest-growing product ISM055 has started clinical research at the end of November this year
.

If the data dimension is the foundation of AI applications and determines the continuous output capabilities of AI in the future, then the products designed by these AI technology platforms are the most intuitive results of AI applications and understand the development stages and challenges faced by the entire field.
Important sample
.

According to Ren Feng, who has a background in biomedicine, the application of AI in pharmaceuticals is still in its infancy compared with relatively mature fields such as AI reading
.
On the one hand, although some AI-driven new drug development projects have been pushed to PCC (preclinical candidate compound), preclinical or early clinical stages, most of these have completed preclinical verification, and none of them have been designed by AI.
The drug completed clinical research and finally proved to be effective in humans
.
On the other hand, there are still a few drugs designed by AI, not many of these innovative products have entered the clinic, and there is a lack of verification on larger-scale products
.

"Combining these two aspects, the depth and breadth of AI verification in the field of new drug research and development is not deep enough
.
"

Being in the early stage and lacking certain verification is the current status of AI new drug development.
The promotion of pipelines and the training and transformation of AI technology in clinical practice have become the top priority of the development of the industry
.

A vice president of a new drug research and development company said that because the product pipeline has very direct value, the industry will be relatively concentrated in this direction
.
However, the application of AI in drug research and development is a big topic, and it is not small to achieve proof of concept (Proof Of Concept)
.
And this link can be broken down into many units and modules to carry out work, such as biomarkers, translational medicine research, or research on finding targets at the source, etc.
Such small modules may make progress faster
.

Regarding the judgment of the development of the industry, Ma Jian believes that it is important to promote the product pipeline, but it is not until the drug is on the market that the application of AI technology is truly implemented and truly successful
.
In fact, it is not necessary to wait until the first drug enters the market stage.
At an earlier stage, such as when a preclinical candidate compound or a drug enters the clinical phase I, the industry's understanding of technology has begun to change, and the tipping point has appeared.

.

When it comes to critical points, it includes thinking about challenges and development directions
.

Ma Jian said that the first thing is to ask the right questions
.
Only by deep coupling with the pharmaceutical industry can we discover valuable, important and profound problems in the industry
.
And professionally correct questions will guide the development of capacity building in a more efficient direction
.
At present, most of the participants in the research and development of AI drugs are early-stage startups, and early-stage startups are better at exogenous AI algorithms, and have not yet formed a sufficiently deep and complete understanding of the industry.
This is the direction that needs to be followed in the future.

.

Another manifestation of the early stage of development of new AI drugs is that the entire field is still undergoing market education, which is also one of the focal points of development
.
According to Ma Jian's observation, market education is also iterating.
At the beginning, the education market was about algorithms, but now it is more about methodology
.

"The market has begun to try and accept AI technology quickly, and the growth rate of customer expansion is very fast
.
And the more companies that try AI, the greater their impact, and they will achieve a certain industry penetration rate and reach a new critical point.
, And then enter the stage of accelerating the application of AI technology in the entire industry
.
"

In the next article, we will approach the participants in the AI ​​pharmaceutical industry and observe their strategies and layout
.