In recent years, AI has been widely used in the fields of bioinformatics data analysis, protein structure elucidation, early drug discovery and disease molecular typing, accelerating the process of new drug research and development and expanding disease diagnosis and treatment solutions. With the rise of new technologies, what are the advances in AI in the research and development of new drugs and the application of disease diagnosis and treatment? What are the challenges? What is the future hold?
On May 26, Shanghai Drug R&D Collaborative Innovation Center and Pudong International Talent Development Center jointly held the 47th Science Café event. Professor Fu Wei of the School of Pharmacy of Fudan University, Chen Rui, Deputy Chief Physician of the Department of Urology, Renji Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, and Dr. Chen Geng, Senior Director of Microbiology, were invited to exchange and discuss the new applications of artificial intelligence in the field of biomedicine.
Overview of the guests' wonderful reports
AIDD and CAD enable new drug discovery
· The current status of the integration of AIDD and new drug research and development
· The current status of integration between CIDD and new drug research and development
● Application of artificial intelligence in prostate cancer diagnosis and treatment
· Research status of diagnosis and treatment of prostate cancer
· Application status of AI in prostate cancer
· Can ChatGPT replace doctors' treatment?
● Application of artificial intelligence in mRNA vaccine research and development
· Application of AI in mRNA vaccine research and development
· mRAN vaccine advantage
· mRNA vaccine development faces the challenge of sequence optimization
· Application of AI in mRNA sequence optimization
· Application of AI in mRNA personalized tumor vaccine development
· SmartNeo analytics platform
01 AIDD and CAD enable new drug discovery
Professor Fu Wei explained the development history, R&D status and respective advantages of AIDD and CIDD in detail from the field of drug design, and shared the team's exploration in the field of AIDD and CIDD.
Compared with traditional drug development, Artificial Intelligence for Drug Discovery (AIDD) and Computer Aided Drug Design (CADD) have obvious advantages, which can combat the inefficiency and uncertainty of traditional drug development. Improve drug development efficiency, reduce R&D costs, and accelerate drug development process.
The current status of the integration of AIDD and new drug research and development
The application of AI technology runs through the whole chain of new drug research and development, including the early drug discovery stage from target discovery, compound synthesis and screening, preclinical research such as crystal form prediction, pharmacological action evaluation, formulation research and development, clinical research stage such as clinical patient recruitment and clinical design optimization, and review and marketing stage such as drug redirection and mass production, all with the participation of AI technology, which can save a lot of time and cost for drug research and development. At present, AI pharmaceutical is still in the climbing stage, mainly focusing on early detection and preclinical stage project development, the therapeutic field is mostly cardiovascular disease and anti-tumor drug field, molecular types include chemical drug small molecules and biological macromolecules, chemical drug small molecules account for more, the degree of innovation is still dominated by me-too/me-better drugs.
◽ The three elements of AIDD
AIDD includes three elements of ABC, namely Algorithm, Computing power and Big data. With the continuous development of computers, computing resources can basically meet the needs of AIDD, and algorithms are constantly developed and optimized, becoming more and more perfect, but big data still restricts the development of AIDD, the reason is that data sources and access are difficult, and the data of many AI pharmaceutical companies involves trade secrets and is difficult to share.
◽ AI competition landscape and capital market
In terms of the number of startups, AI pharmaceutical startups continue to increase. As of 2019, the number of AI pharmaceutical companies in the United States ranks first in the world, and China ranks fifth. At present, the number of domestic startups has increased from five in 2019 to more than a dozen, and these companies run through the whole chain of drug research and development, involving various research fields.
From the perspective of the capital market, after 2020, the global AI capital market entered a financing explosion period, from 2018 to 2021, the annual growth rate of financing amount was 110%, and the global AI pharmaceutical investment and financing scale reached 3.28 billion US dollars in 2021, an increase of 2.5 times year-on-year. Mainland AI pharmaceutical started slightly lagging behind overseas, and a small number of financing events began to occur since 2015, and 2020 was a turning point in the mainland AI pharmaceutical field, with the number of financing projects doubling and the total financing increasing by about 7 times year-on-year. In 2021, the capital market reached 6.1 billion yuan.
◽ The application of AI technologies such as ChatGPT in the field of new drug research and development
ChatGPT is a natural language processing model based on machine learning and neural networks. ChatGPT's way of thinking is mainly based on big data and algorithms for reasoning and learning, while the human brain is thinking and cognition through the coordination between neural connections and brain regions. In recent years, there are many products based on ChatGPT-assisted drug development, mainly including: Pangu developed by Huawei, which is a drug molecular model with a 20% higher accuracy rate than traditional methods; Microsoft-developed BioGPT, which is pre-trained on a huge corpus of more than 15 million abstracts from PubMed, and the accuracy of model prediction in PubMedQA's detection is as high as 81%; In addition, Profluent has developed ProGen software with protein design capabilities to design novel functional proteins.
The current status of integration between CIDD and new drug research and development
◽ The basic process of CADD
CAD is the process of studying the interaction patterns between chemical small molecules and biological macromolecules, and then conducting structure-based drug design to discover active small molecules. From the small molecule compound database and protein database, the drug molecules and target proteins were screened out, so that they interacted to form complexes, and the structure of the complex was determined by X-Ray, NMR, electron microscopy and other technical means and calculated by quantum chemistry, molecular mechanics, and molecular dynamics, and the structure of small molecule compounds and proteins was continuously optimized and designed, and the three-dimensional structure of a small molecule and protein complex was obtained, and then the cycle test of active structure-activity relationship and pharmacokinetic/pharmacodynamic structure-activity relationship was carried out. The result was a preclinical candidate compound.
◽ Traditional CAD application scenarios and limitations
The application scenarios of traditional CAD include protein structure prediction, small molecule structure construction, small molecule-protein interaction, free energy computing, structure-based drug design (SBDD), pharmacophores, nature-based drug design (PBDD), ADME/T, protein conformation, protein structure and function relationship, etc. Traditional CAD runs through all aspects of early drug discovery, but because traditional CAD is based on static structure for drug molecule design, and drug and target interaction is a dynamic process, there are conformational changes, CAD does not take into account the conformational changes of small molecules and protein formation complexes, that is, insufficient consideration of protein flexibility, therefore, the accuracy of CAD prediction is limited.
◽ The team explores the application of CAD technology in the development of new drugs
In order to solve the limitations of CADD, Professor Fu's team developed dynamic virtual screening technology to study the dynamic conformational changes of proteins through molecular dynamics simulation, using a series of computational techniques, including active site detection and cluster analysis, and establishing multiple pharmacophore models for lead compound discovery. This dynamic virtual screening method has carried out the design and discovery of a variety of drugs, including the design and discovery of A-GPCR drugs, and the design and discovery of analgesic drugs in the field of central nervous system diseases. Drug design and discovery targeting nuclear receptors, drug research and development in the field of anti-inflammatory and immune diseases; He also conducts research in the field of molecular design and mechanism of drug delivery systems and urease biopharmaceuticals.
G protein-coupled receptor accounts for about 45% of the drug target, human taste, vision, smell and other senses and pain are related to the receptor, the team used CAD technology to screen from 96 compounds and found more than 20 active compounds, based on these compounds did a series of modifications, docking the central nervous system dopamine receptor and serotonin receptor subtypes, found that the skeleton of some of the active molecules is the molecular structure of anti-tumor drugs, However, it is also a binder of serotonin-2A (5-HT2A) and has multi-target and antagonistic characteristics. On the basis of these studies, the team developed 5-HT1AR agonists, adopted dynamic virtual screening technology, built multiple pharmacophore models through large-scale kinetic simulation, designed and discovered a series of active molecules, and finally screened out 10 new structural compound molecules with functional activity up to 7 nmol after optimization to discover new structural drug candidate molecules with anti-schizophrenic and antidepressant efficacy. In addition, the team also used CAD technology to develop powerful analgesic drugs, and found more than 300 highly active molecules, of which PCC molecules are 50-100 times more active than morphine, and have strong cancer pain analgesic effect, can be used as a broad-spectrum analgesic drug, this agonist has obtained 7 patents. Up to now, the team has published more than 40 articles in the field of GPCR, authorized 13 patents, and 1 patent has entered the US market.
Future outlook: AIDD and CAD complement each other
AIDD is a large-scale training model, more dependent on algorithms and computing power, insufficient consideration of the active structure of drug molecules, while CAD is more inclined to the analysis of drug structure-activity relationship, using the concept of pharmacophores and fragment-based drug design, AID and CAD have their own advantages, Professor Fu's team adopts the new drug research and development ideas of CAD and AID fusion, on the basis of AIDD's graph-based molecular characterization method, integrates the concept of CAD based on the molecular generation of pharmacophores and active fragments. Carry out drug molecular design and apply it to the development of analgesic drugs in the field of central nervous system. Professor Fu said that the future development direction of AIDD and CAD is to integrate and complement each other, so as to better empower the development of new drugs.
02 Application of artificial intelligence in the diagnosis and treatment of prostate cancer
Starting from the current research status of prostate cancer diagnosis and treatment, Dr. Chen Rui introduced the challenges of prostate cancer diagnosis and treatment, and shared the application of AI in prostate cancer imaging diagnosis, pathological diagnosis, construction of patient molecular typing, and the team's research on analyzing multimodal physical examination data to assist early diagnosis through artificial intelligence models.
Research status of diagnosis and treatment of prostate cancer
◽ Prostate cancer epidemiological data
Prostate cancer is the second most common malignancy among men worldwide. Although the incidence of prostate cancer in mainland China was not high in the past, in recent years, its incidence has increased year by year, and has jumped from thirteenth to sixth, with an annual growth rate of 13%. With the popularization of tumor screening in mainland China and the intensification of aging trend, prostate cancer will become a major disease affecting the health of elderly men.
◽ Challenges in prostate cancer diagnosis and treatment
At present, the diagnosis and treatment of prostate cancer in mainland China has been at the forefront of the world, the minimally invasive surgery technology of Dr. Chen's team has been at the international first-class level, da Vinci robot, domestic robot and other technologies are constantly updated and iterated, and the surgical treatment effect is getting better and better.
However, the diagnosis and treatment of prostate cancer still faces two difficulties: the first is early diagnosis. First, early prostate cancer lacks symptoms, and many patients are difficult to detect until metastasis. Secondly, different patients have different degrees of tumor differences, some tumors with a high degree of malignancy and rapid progression are the most in need of early diagnosis and timely treatment, while some tumors have a low degree of malignancy and can be actively monitored. So how to diagnose highly malignant prostate cancer is a challenging problem. Dr. Chen's team explored new methods of research and diagnosis, and proposed diagnostic strategies that matched Chinese patients by "sinicizing" PSA indicators developed in European and American populations, and also developed new molecular diagnostic markers to improve the diagnostic effect. The second challenge is late-stage treatment. After prostate cancer progresses to an advanced stage, the treatment options are very limited, and there is an urgent need to molecularly type drug-resistant advanced prostate cancer patients to find suitable treatment drugs in a targeted manner. Dr. Chen's team is also committed to advanced molecular typing and precision treatment of prostate cancer, and has proposed related treatment strategies, and the efficacy has been improved, but there is still a gap from satisfactory treatment effect.
Application status of AI in prostate cancer
In February 2023, a medical-engineering fusion team composed of scholars from Changhai Hospital, Renji Hospital and East China University of Science and Technology discussed the application of artificial intelligence in the diagnosis and treatment of prostate cancer, and combined with the team's research results, compiled the book "Application of Artificial Intelligence in the Diagnosis and Treatment of Prostate Cancer". At present, the most mature application of artificial intelligence in the field of prostate cancer is pathological diagnosis and imaging diagnosis of prostate cancer.
◽ Application of AI in prostate cancer imaging diagnosis
Many patients have elevated PSA, but not necessarily all have prostate cancer, and blindly piercing all patients will bring unnecessary trouble to patients without tumors. In response to such a situation, radiologists have developed multi-parameter magnetic resonance, which can simultaneously obtain information from multiple weighted images, indicate suspicious areas of prostate tissue, and then perform puncture biopsy according to suspicious areas, known as "magnetic resonance-guided targeted puncture", and this precise method has been used clinically. However, the number of experienced prostate radiologists is very limited, usually concentrated in large cities and large hospitals, resulting in many small and medium-sized cities and small hospitals unable to complete the reading results well. At present, a team has applied AI to the identification of tumor lesions in magnetic resonance images and obtained good results, and it is expected that this technology will soon enter the decision-making of prostate puncture by clinical auxiliary radiologists and urologists.
◽ Application of AI in pathological diagnosis of prostate cancer
The application of digital pathology and artificial intelligence is constantly changing the way pathology is diagnosed. The digital slide scanner can digitize pathology slides, a thumb-sized pathology section, which can be enlarged and scanned into clear panoramic pathology scan slides with a data volume of up to 2-3GB. By allowing artificial intelligence to learn the manually sketched tumor area and training the artificial intelligence model to help pathologists screen out the tumor area, there are also relatively mature models that can reduce the burden of pathologists and will also be put into clinical use in recent years.
◽ Explore AI-assisted molecular typing of prostate cancer patients
The molecular typing of breast cancer has been at the forefront of a variety of tumors, and doctors can take different treatment plans according to the different molecular types of patients. In contrast, the molecular typing of prostate cancer is still in its infancy, and the scientific research team tries to use artificial intelligence, combined with the big data of existing clinical patients, to explore the study of molecular typing of prostate cancer. At present, preliminary research results have been achieved, and it is expected to be applied to clinical diagnosis and treatment decisions in the future.
◽ Artificial intelligence analyzes multimodal physical examination data to realize early warning of people at high risk of prostate cancer
With the aging of the population, the burden of prostate cancer screening in mainland China will continue to increase, if routine physical examination information can be used as the basis for prostate cancer screening, it is expected to reduce unnecessary punctures, reduce the burden of medical institutions, and reduce national medical costs. Dr. Chen's team analyzed the clinical data of thousands of people in hospitals including Shanghai, Nanjing, Seoul, Hong Kong and other regions, and used XGBoost artificial intelligence algorithm to build an interpretable diagnostic prediction model, which can reduce unnecessary punctures by 38% while ensuring detection rate.
Can ChatGPT replace doctors' treatment?
This year, the JAMA Journal of Internal Medicine reported on a study that tested the performance of ChatGPT by collecting questions from foreign doctor-patient consultation websites, and the results showed that compared with doctors' online responses, the results showed that ChatGPT scored even better than doctors' online responses.
At the same time, Dr. Chen's team also published research on the application of ChatGPT in prostate cancer consultation, consultation and medical teaching in the international translational journal Translational Journal of Medicine. The team designed 22 questions that patients may ask related to the diagnosis and treatment of prostate cancer, and the researchers found that ChatGPT can give good answers to most of the questions, especially the answers to routine questions such as prostate cancer stage, grade, and conventional treatment are very accurate; However, for the challenging questions that require a comprehensive analysis of the patient's condition, ChatGPT cannot answer it well.
The study evaluated a variety of mainstream language models such as ChatGPT, Perplexity, and YouChat, and the researchers found that ChatGPT is the highest-scoring model, with higher accuracy, better humanistic care, and stronger stability than other models. Dr. Chen believes that ChatGPT cannot replace doctors at present, but ChatGPT can be used to better assist diagnosis and treatment. In addition, based on legal and ethical considerations, artificial intelligence cannot bear the consequences and legal liabilities caused by mistaken clinics, which is also a problem that needs to be considered when applying artificial intelligence in the medical industry.
Outlook for the future
How will artificial intelligence help doctors improve the diagnosis and treatment of diseases in the future? It is envisaged that the artificial intelligence system can network the patient's health information and screening center, analyze the patient's data, including routine physical examination information, living habits, and tumor markers, determine whether the patient is at risk of disease, and whether further examinations such as puncture pathology are required. For patients who need to be pierced, artificial intelligence can assist doctors in image analysis; For patients who need surgery, artificial intelligence can assist doctors in carrying out robotic surgery; Artificial intelligence can also be used to optimize the follow-up plan after surgery and improve the follow-up effect. Finally, Dr. Chen said that facing many challenges in the diagnosis and treatment of prostate cancer, researchers in the fields of more information engineering and pharmacy are needed to work together to overcome the problems.
03 Application of artificial intelligence in mRNA vaccine research and development
Dr. Chen Geng introduced the application of AI in the research and development of mRNA vaccines from a technical level, including the optimization of mRNA sequences, the development of personalized tumor vaccines and the construction of a SmartNeo analysis platform for the identification of tumor neoantigens.
Application of AI in mRNA vaccine research and development
Artificial intelligence technology can play a very important role in the field of mRNA and can be applied to various mRNA related fields. For example, in the upstream important links of each mRNA vaccine R&D pipeline, AI can perform target screening based on bioinformatics and optimize mRNA sequences.
mRAN vaccine advantage
Compared with traditional inactivated vaccines, mRNA vaccines have a short development and production cycle, good safety, and easy scale-up production
mRNA vaccine development faces the challenge of sequence optimization
The development of mRNA vaccines basically involves sequence optimization. Because the development of mRNA vaccines is usually based on antigen design, antigens are amino acid sequences, and most amino acids can be encoded by at least two codons. A complete mRNA structure usually contains a cap and non-coding region (UTR) at the 5' end, and a UTR and Poly(A) tail at the 3' end, with the coding region encoding the antigen in the middle. Therefore, each element of mRNA may affect the translation efficiency and stability of mRNA. Converting antigens into mRNA sequences for encoding presents many challenges due to the complex correspondence between codons and amino acids. For example, the human body has 64 codons, but these codons only encode 20 amino acids, so one amino acid may correspond to multiple codons, and each amino acid chooses which codon to encode is very particular. Take the new crown's spike protein as an example, it has more than 1200 amino acids, if the spike protein is encoded by codons, the coding amino acid sequence has 10 to the power of 632 possibilities, it is very challenging to choose which mRNA sequence is good, artificial intelligence technology can help screen and optimize mRNA sequence.
Application of AI in mRNA sequence optimization
Codon selection has been shown to affect the stability and translation efficiency of mRNA sequences. Because some codons are used more frequently in cells, but some codons encoding the same amino acid may be used less frequently, choosing to use more frequent codons can potentially increase the speed of mRNA translation into antigens. However, it is not possible to choose to use frequent codons to encode amino acids, which may cause instability of mRNA sequences, and sequence design needs to consider many factors.
Siwei collaborated with Baidu to develop an AI algorithm called LinearDesign for mRNA sequence optimization and design, and the results have just been published in Nature. This algorithm is based on the comprehensive consideration of mRNA minimum free energy (MFE) and codon adaptation index (CAI) to design and optimize mRNA sequences. Theoretically, the more stable the secondary structure of mRNA, the longer it takes to be expressed in the cell, and the codon adaptation index reflects the speed at which mRNA is translated into protein. If mRNA translation is fast and stable, the protein expression can be higher. Therefore, it is possible to design the ideal mRNA molecule by considering these two important factors together. The results of this Nature article show that the sequence of the new crown mRNA vaccine optimized by the LinearDesign algorithm can not only improve the stability of mRNA and protein expression, but also the vaccine can induce higher levels of binding antibodies, neutralizing antibodies and cellular immunity in vivo. At the same time, in the test chickenpox and herpes zoster mRNA sequences, it was shown that the optimized mRNA sequences also had higher stability and protein expression levels, and induced the production of higher binding antibodies in vivo.
Obviously, artificial intelligence can play a very large role in the design and optimization of mRNA sequences, and can quickly design mRNA sequences with high expression and good stability, which greatly promotes the research and development of mRNA vaccines.
Application of AI in mRNA personalized tumor vaccine development
◽ mRNA personalized oncology vaccine development process
Usually the patient's tumor tissue sample is first obtained for sequencing, the tumor neoantigen is identified, the corresponding mRNA vaccine is designed, and then the mRNA vaccine is injected into the patient's body, and the antigen-presenting cells in the human body take these antigens, including DC cells or B cells, present the antigen to T cells, activate CD8 cells and CD4 T cells in the patient's body, and then CD8 and CD4 T cells recognize the specific antigens presented on the surface of the tumor, and kill or destroy these tumor cells accordingly.
Identification of high-quality neoantigens with immunogenicity is key upstream in the development of mRNA personalized vaccines.
At present, studies have shown that tumor neoantigens are produced in a variety of ways, including antigens produced by various mutations at the genome level, variable cutting or editing at the RNA level, and some non-classical translations and protein post-translational modifications. Moreover, studies have shown that different types of tumors, their mutational burden and the frequency of tumor neoantigen production are different, compared with other cancers, melanoma, lung cancer and gastric cancer and other tumors have a higher frequency of tumor mutation frequency and neoantigen production frequency, these patients may be easier to screen out more neoantigens. In addition, even within the same cancer, the frequency of tumor neoantigen production varies greatly between different individuals. Therefore, designing a good personalized mRNA tumor vaccine requires the identification of high-quality neoantigens with immunogenicity.
Identifying high-quality neoantigens is like finding a needle in a haystack, because each tumor patient's tumor cells can produce thousands of mutations that can change the protein sequence, and relevant studies have shown that the proportion of peptides containing mutations that can be bound by the body's MHC may be only about 1%, only about 0.5% may be recognized by T cells, and only about 0.3% may be able to effectively process and present antigens in tumor cells, and fewer will eventually activate the immune response of T cells in the patient. Therefore, the identification of truly immunogenic novel antigens requires overcoming many difficulties. Therefore, researchers can use artificial intelligence to help identify and screen high-quality neoantigens.
SmartNeo analytics platform
Dr. Chen led the team to develop a SmartNeo platform that can identify high-quality tumor neoantigens. Based on the patient's multi-omics sequencing data, the platform can perform data quality control, identify unique mutations in patients' tumor tissues, analyze patients' HLA typing, obtain gene and transcript expression patterns in tumor tissues, and the characteristics of mutations in different omics dimensions, and finally identify and screen high-quality antigens HLA-I and HLA-II neoantigens based on corresponding AI methods and models, forming a one-stop analysis platform.
In order to test and analyze the performance of the SmartNeo platform in identifying high-quality neoantigens, the standard test data set of TESLA (Tumor Neoantigen Selection Alliance) published in cell in 2020 was used as a benchmark, and compared with other neoantigen analysis platforms reported in recent years, it was found that the analysis speed of the SmartNeo platform can be at least twice as fast as that of other platforms. And more high-quality neoantigens can be identified. Collecting the results of clinical trials carried out at home and abroad, it shows that the proportion of antigens predicted by the SmartNeo platform to cause immune response is also higher than that reported in other articles. Overall, the SmartNeo analytics platform has advantages over other platforms.
summary
Artificial intelligence is playing an increasing role in the field of biomedicine, which can not only reduce R&D costs and shorten R&D cycles; In terms of sequence optimization in the mRNA field, AI can quickly identify and screen sequences with higher expression and better stability. In the research of mRNA personalized tumor vaccines, AI can help identify high-quality neoantigens and analyze sequencing big data more efficiently.
Science Café
Science café is a cross-border coordination and cooperation platform jointly organized by Shanghai Drug R&D Collaborative Innovation Center and Pudong International Talent Development Center, which integrates information exchange, brainstorming and creative integration. One session and one meeting, a cup of coffee, gather collective wisdom, and discuss the frontier topics of biomedical innovation from multiple angles. Since 2014, more than 40 brand salon activities have been held.
If you are interested in the frontier topics of biomedical innovation or have the intention to cooperate, please contact us in the following ways, and look forward to working with you.
Contact: Ms. Ling
Contact number: 58336301, 15000788873
end.
Article and picture source: Shanghai Drug R&D Collaborative Innovation Center public account (changed)
If there is any infringement, please contact Xiaobian in the background to delete