Humans created AI, can AI in turn save humanity? In addition to Q&A chat painting, ChatGPT is accelerating its penetration into subdivided fields, and AI pharmaceuticals are becoming the pioneers of AI applications.
Tech giants cross over into the medical track
The value of the medical track has made many technology giants restless, and the iteration of AI technology represented by ChatGPT has finally allowed technology companies to see the opportunity to knock on the door of the medical field.
OneZero introduced Huawei's Pangu model yesterday, which has a layout in the medical subdivision track, and the mid-term Pangu drug molecule model is itself a large model jointly trained by HUAWEI CLOUD and the Shanghai Institute of Materia Medica, Chinese Academy of Sciences, which can realize AI-assisted drug design for the whole process of small molecule drugs.
Domestic IT giants layout AI pharmaceutical data platform Source: Yiou Think Tank
In addition to Huawei, Alibaba has medical cloud AI, which is used in fields such as gene sequencing, thyroid nodule recognition, lung nodule recognition, and digital simulation clinical experiments. Tencent specializes in cloud deep intelligence drugs, which are used in drug development processes such as protein structure screening. Byte has AILab, based on AI algorithms, to support the discovery and manufacture of drugs under investigation.
According to media reports, Baidu Wenxin Yiyan's first product GBI-Bot in the pharmaceutical industry was officially released recently, this pharmaceutical pendant dialogue robot, using Baidu Lingyi Zhihui's technology accumulation in the medical and health industry, realizes the organic combination of Wenxin Yiyan and GBI professional database.
In addition to mainland technology companies actively entering the medical field across borders, global technology giants such as NVIDIA and Microsoft are also actively cooperating in the field of AI pharmaceuticals.
At last week's Nvidia Spring GTC conference, Nvidia said it would collaborate with Mitsui on the Tokyo-1 project, which aims to improve Japan's pharmaceutical capabilities through technologies such as high-resolution molecular dynamics simulations and generative artificial intelligence (AIGC) models for drug discovery. Japanese pharmaceutical companies and startups are expected to participate in the Tokyo-1 project, and Mitsui & Co.'s subsidiary Xeureka will run Tokyo-1 and officially launch the project later this year.
Previously, Microsoft and Novo Nordisk teamed up to combine Microsoft's computing services, cloud and artificial intelligence (AI) with the latter's drug discovery, development and data science capabilities. Novo Nordisk also said in a September 12 press release that under the partnership, Microsoft will provide artificial intelligence technology, basic science models and expertise, and work with Novo Nordisk's data scientists and experts in early-stage research and development to accelerate pharmaceutical research and development.
In recent years, AI has accelerated the development of new drugs, and has participated in almost the entire process from drug target discovery to clinical trials.
Wen Wen, founder of Huanyi Biotechnology, mentioned that more than 190 new drugs successfully approved by the FDA in the past 10 years have basically been developed by more than 120 companies, and even the world's top 10 pharmaceutical companies rely on the acquisition of small pharmaceutical companies, "which fully shows that the efficiency of drug research and development is extremely low and has not formed a scale effect." ”
At its core, the drug discovery process is a data and engineering problem. In sharp contrast to the anti-Moore's law of pharmaceuticals is "computing power", AI that has blossomed everywhere in many fields of autonomous driving, the greater the computing power, the lower the marginal cost, AI and other computing into pharmaceutical is almost logical.
Compared with the traditional drug development model, AI drug development has the advantages of shortening the development cycle, saving capital costs and improving the success rate. According to Drug Times, drug development in the traditional mode takes 4-5 years in the preclinical stage, while the new drug development pipeline based on AI and biological computing can complete preclinical drug development in an average of 1-2 years.
AI that explores the forbidden area of life
Technology has played a significant role in the development of drugs from "accidental discovery" to the creation of new drugs through the design of drug molecules.
There are two technical models for research and development with the help of computation - one is an artificial intelligence method based on computing power, algorithms, and data, and the other is a high-precision computational chemistry method based on a physical model. The former essentially uses models such as machine learning to allow AI algorithms to find a way to solve problems in continuous optimization iterations, and then optimize the whole process of drug development; The latter is to calculate the interaction force between drug molecules and target molecules from the level of microscopic particles such as molecules and atoms, and use AI to improve the speed and accuracy of computing.
At present, the application of AI in the field of medicine is roughly divided into three stages, the first stage is AI image recognition, the use of computer vision, deep learning and other artificial intelligence technologies, endoscopy, mammography, ultrasound, CT, MRI, pathology, fundus photography, OCT and other medical images for learning and training, can effectively assist doctors in diagnosis and early screening of major diseases and other tasks.
The second stage, represented by AlphaFold, trained AlphaFold using nearly 170,000 different protein structures in a protein database, as well as a database of protein sequences containing unknown structures. Through continuous iteration, the AlphaFold system learned the ability to accurately predict protein structure based on amino acid sequences.
The third stage is to use the current "OpenAI" class semantic recognition system (such as ProGen) to create structures that are not seen in nature through simple instructions. ProGen iteratively optimizes by learning the probability of predicting the next amino acid given past amino acids in the original sequence, without clear structural information or paired coevolution assumptions.
The leading player in AI drug discovery is the United States. A large number of medical AI companies in the United States are concentrated in the field of drug discovery, and more than 50% of the world's AI drug discovery companies are concentrated in the United States. Schrodinger and RelayTherapeutics are the first companies to go public in the field of AI drug discovery in the United States.
In November 2020, AlphaFold2, owned by Google, DeepMind, solved a biological puzzle - protein folding. In July 2021, Google and the European Laboratory for Molecular Biology (EMBL) used AlphaFold2 to predict the three-dimensional structure of 350,000 proteins based on amino acid sequences, covering almost the approximately 20,000 proteins expressed in the human genome.
This has made all walks of life pay attention to the application of AI in the field of drug research and development, and since then, the financing activities of AI pharmaceutical research and development have become more and more frequent.
AI pharmaceuticals have gradually become the darling of capital
Start-ups are the main force of AI pharmacy, and they are also the main promoters and practitioners of AI pharmacy.
Data show that as of Q3 2022, there are about 600 AI drug discovery companies in the world, a year-on-year increase of 21.6%. Among them, the number of AI pharmaceutical companies in the United States ranks first in the world, with 343, an increase of nearly 3 times in two years; There are also nearly 80 AI pharmaceutical-related companies in China.
Global AI Pharma Financing, Source: BiopharmaTrend
Although the domestic AI pharmaceutical started late, its strength should not be underestimated. The McKinsey report shows that Chinese AI drug development companies are on the rise, taking the top 20 AI pharmaceutical companies in the world for pre-IPO financing, accounting for one-third of the number of companies from China.
As the intersection of the field of artificial intelligence and the pharmaceutical field, AI drug research and development has been encouraged by the state for innovative drug research and artificial intelligence development, and has also made certain progress.
Not long ago, AI+ synthetic peptide new drug research and development enterprise Chengyuan Technology (Syneron Tech), announced the completion of tens of millions of dollars in pre-A round of financing, co-led by Lenovo Venture Capital and Gree Production, among which, as a state-owned asset in Zhuhai, Gree Gold Investment has attracted a lot of attention to AI pharmaceuticals, and capital investment itself is also an attitude.
According to statistics, the average cost of launching a new drug has increased from US$1.188 billion in 2010 to US$2.508 billion in 2022, while the global return on investment on new drugs has fallen from 10.1% in 2010 to less than 2% in 2018.
The news of the suspension of the first domestic new crown drug brushed the screen last night, spending 1.3 billion yuan on research and development but only selling 50 million yuan, which dragged the market value of the company into the abyss.
In contrast, AI pharmaceuticals can optimize processes and reduce labor costs by cross-aligning data, accelerating screening, generating from scratch, discovering targets and new drugs faster and cheaper, and improving the success rate of drug design and screening through high-throughput trial and error.
However, due to the high technical threshold of artificial intelligence, it is difficult for pharmaceutical companies to have technical advantages in the short term, so they will choose to cooperate with technology giants. Technology companies are responsible for building algorithm models, while pharmaceutical companies are responsible for using data to train models, which gives technology giants such as Baidu, Alibaba, and Huawei the opportunity to enter the medical field across borders.
Of course, pharmaceutical research and development is a long cycle, a large amount of investment, even if artificial intelligence gradually penetrates into various pharmaceutical links, many decisions still need the help of pharmaceutical experts, and the time required is only relatively short. Therefore, for enterprises, strong financial resources are the basis for supporting their early research and development. For domestic AI pharmaceutical companies, the biggest problem today is still how to get orders from pharmaceutical companies, compared to the United States, except for XtalPi's orders from Pfizer, almost no companies can get orders from large pharmaceutical companies.