Cell Analysis | Molecular Mapping | IND
Biometrics | Gene Sequencing | AlphaFold
Against the backdrop of the rapid development of cell analysis, visual recognition, biometrics, gene sequencing, IND, and AlphaFold, various emerging technologies that benefit mankind have begun to emerge. Recently, the 20th Anniversary Theme Summit of "Top Ten Breakthrough Technologies" of MIT Technology Review was successfully held in Future Science and Technology City, Yuhang District, Hangzhou.
Yang Wei, Academician of the Chinese Academy of Sciences, Chairman of the Development Committee of Zhejiang University, Professor of Zhejiang University, He Kebin, Academician of the Chinese Academy of Engineering, Professor of the School of Environment, Tsinghua University, Zhang Yaqin, Academician of the Chinese Academy of Engineering, Dean of the Intelligent Industry Research Institute (AIR) of Tsinghua University, etc. Scientists, industry leaders, and business elites were invited to attend.
The breakthrough technologies released this time include: new crown oral medicine, practical fusion reactor, termination code, AI protein folding, PoS authority certificate, long-term grid energy storage battery, AI data generation, malaria vaccine, carbon removal factory and new crown pneumonia mutation tracking .
new crown oral medicine
Since the global outbreak of the new crown epidemic, major pharmaceutical companies and scientific research institutions around the world have been working hard to develop vaccines that can effectively prevent new crown virus infection and effective treatment drugs for new crown virus.
According to statistics, as of the beginning of May, there are more than 1,200 COVID-19 drugs under research worldwide, and nearly 50% of the projects are in the IND and above R&D stages. 15 applications for marketing have been submitted, involving more than 1,000 companies; there are currently more than 50 drugs in the world (including vaccines) have been approved for COVID-19 indications, including 12 small molecule drugs and more than 30 biological drugs.
From the perspective of the mechanism of action, among the approved small molecule chemical drugs, there are 3 RdRp inhibitors (RNA polymerase inhibitors since RNA), namely Gilead’s Remdesivir, Toyama Chemical’s Favipiravir, Merck’s molnupiravir; in addition, Pfizer’s 3CL protease inhibitor Paxlovid, COVID19 replicase polyprotein 1a inhibitor combination drug Nematevir + ritonavir, Incyte’s JAK inhibitor baricitinib, etc.
In the category of biological drugs, the number of drugs whose mechanism of action is COVID19 spike glycoprotein modulator accounts for the highest proportion (about 50%); from the perspective of therapy type, except for vaccines, neutralizing antibodies account for the majority, including sotrovimab, Cassie Reimumab + Idulumab, Barnavirumab, etc. In addition, nucleic acid drugs such as tozinameran, elasomeran, and ZyCoV-D have also been approved for the treatment of new crowns.
At present, there are two oral small-molecule specific drugs for the new crown that have been marketed in the world, Merck's Molnupiravir and Pfizer's Paxlovid. A total of more than 10 new crown drugs (including vaccines) have been listed in China, including oral drugs favipiravir, nematevir + ritonavir, neutralizing antibody ambavirumab + romisvirumab, etc., most of which Emergency Use Authorization/Conditional Approval.
Compared with remdesivir and neutralizing antibody drugs that require injection, oral small molecule drugs have more advantages, and have therefore become a popular track for global new crown drug development.
Advantages of new crown oral small molecule specific drugs:
1. Patients have high tolerance and good compliance, which is convenient for inhibiting the proliferation of the virus in the early stage of infection and avoiding severe disease
2. The price is low. The price of mupinavir in the United States is 700 US dollars per person, which is only one-third of the neutralizing antibody
3. Easy to transport and distribute. Compared with antibody drugs that require intravenous injection, oral small-molecule drugs are undoubtedly much more convenient. In underdeveloped countries with severe epidemics and backward medical conditions, oral small-molecule antiviral drugs are more practical
Judging from the competition situation of domestic enterprises, there are currently more than 100 domestic enterprises participating in the research and development of new crown drugs, involving more than 150 research and development projects. There are more than 10 new crown oral drugs under development in China, including Azvudine, VV116, Proxalutamide, SIM0417, RAY003, etc. The listed companies involved include Junshi Biotechnology, Kintor Pharmaceuticals, Simcere Pharmaceuticals, Zhongsheng Pharmaceuticals, etc. .
From the perspective of drug research and development progress, there are 6 drugs in the clinical trial stage, of which the three fastest progressing drugs are Azvudine from Real Bio, VV116 from Junshi Biotech, and proxalutamide from Kintor Pharmaceutical. The first domestically produced oral small-molecule specific drug for the new crown is basically locked in these three drugs.
1. Research and development of "overspeed" VV116
VV116 is a novel oral nucleoside anti-SARS-CoV-2 drug, which is an RdRp inhibitor and can inhibit the replication of SARS-CoV-2. At present, it has obtained emergency use authorization in Uzbekistan. This is another new crown oral drug approved for marketing in the world after the approval of Merck and Pfizer new crown oral drugs. It is jointly developed by Shanghai Institute of Materia Medica, Chinese Academy of Sciences, Wuhan Institute of Virology, Chinese Academy of Sciences, Xinjiang Institute of Physics and Chemistry, Chinese Academy of Sciences, Wangshan Wangshui Biomedical Co., Ltd., and Central Asia Drug Research and Development Center, Chinese Academy of Sciences.
According to the news released by Junshi Biotechnology, VV116 is used in the early treatment of patients with mild to moderate new coronavirus pneumonia with high risk of progression to severe disease including death The phase III registration clinical study (NCT05341609) reached the primary endpoint and secondary efficacy endpoint preset by the protocol. And issued an announcement saying that it will communicate with the drug regulatory authorities in the near future to submit a new drug listing application.
After the release of its research results, discussions and doubts about the VV116 clinical trial in the industry also flocked. The reason is that the research and development process of its clinical trials is too fast, and the industry believes that its research is directly transferred to Phase III clinical trials, which is not strictly compliant. shorter. In addition to the "speeding" of the research and development process, VV116 has also caused some controversy in terms of safety and patents.
Although controversial, Junshi Biosciences stated that the study has reached the primary endpoint of the clinical protocol. At least in terms of the duration of clinical recovery, the efficacy of VV116 is no less than that of PAXLOVID. In general, the experimental data released successively has boosted the market's confidence in the follow-up development of VV116.
2. Azvudine, who has not been approved first
Azvudine was originally a drug for the treatment of AIDS. In July 2021, it was approved for marketing in combination with nucleoside reverse transcriptase inhibitors and non-nucleoside reverse transcriptase inhibitors to treat adult HIV-1 infected patients with high viral load, becoming my country's first truly independent intellectual property rights anti-HIV drugs.
After the outbreak of the new crown epidemic, Real Biology carried out research on the treatment of new coronary pneumonia with azvudine. From the point of view of the target, Azvudine acts on RdRp (RNA polymerase), which is different from Pfizer’s Paxlovid on 3CLpro (3C-like protease), but similar to Merck’s Molnupiravir.
In mid-April, Real Biology released some data on Azvudine's anti-coronavirus use. Judging from the results of Phase II clinical trials and part of Phase III clinical trials, the time for Azvudine nucleic acid to turn negative is 3-4 days, the average medication time is 6-7 days, and the average discharge time is 9 days. The treatment effect of severe and mild cases is similar, and it is also effective for patients who are ineffective with other drugs, and unlike Paxlovid, it needs to be taken in the early stage of infection with the new crown.
On May 12, Junshi Biotechnology disclosed the price for the first time. According to media reports, the price of VV116 in Uzbekistan is US$185, or about RMB 1,243. In the case that the results of anti-new coronavirus clinical trials have not yet been released, Real Biology has finalized three production and distributors (China Resources Double Crane, Xinhua Pharmaceutical, Aoxiang Pharmaceutical) for Azvudine, so it is also jokingly called "one drug" by the industry. Women marry three times".
3. Proxalutamide with twists and turns
Proxalutamide was originally a second-generation AR antagonist used by Kintor for the treatment of prostate cancer. After the outbreak of the new crown, clinical trials confirmed that the drug has a therapeutic effect on the new crown. At the beginning of 2021, in the Phase III clinical trial in Brazil, proxalutamide can reduce the risk of death of patients with severe new coronary pneumonia by 92%, which was once regarded as "the hope of the people". Later, because the interim analysis of the phase III clinical trial did not reach statistical significance, it caused a lot of controversy.
In April 2022, Kintor Pharmaceuticals announced the key data of the clinical phase III trial, pointing out in particular that "proxalutamide effectively reduces the hospitalization/mortality rate of patients with new crowns, especially for all patients who have taken the drug for more than 7 days, and those with high risk It is statistically significant that the protection rate of middle-aged and elderly patients with COVID-19 has reached 100%, from clinical failure to 100% effective, it is worth looking forward to whether proxalutamide can make a comeback in the competition.
In addition, domestically-made new crown drugs that are still in the early stages of research and development include FB2001 of Frontier Bio-U (688221.SH), SIM0417 of Simcere Pharmaceuticals (02096.HK), and Ascletis Pharmaceuticals-B (01672.SH). HK)’s ASC10 and ASC11, Guangsheng Zhonglin/WuXi AppTec (603259.SH)’s 3CL protease inhibitor, Zhongshengruichuang’s RAY003, etc.
practical fusion reactor
Researchers at Commonwealth Fusion Systems slowly charged a 10-ton D-shaped magnet and increased the field strength until it exceeded 20 Tesla (T). This is a new record for a magnet of its kind. The company's founders say the feat solves a major engineering challenge in developing a compact, inexpensive fusion reactor.
Fusion power generation has been a dream of physicists for decades. At temperatures well above 100 million degrees Celsius, as in the sun, nuclei fuse together, releasing enormous amounts of energy in the process. If researchers can achieve these reactions in a controlled and sustained manner here on Earth, it could provide a cheap, continuous, carbon-free source of electricity, using a virtually unlimited source of fuel.
In one approach, magnets are used to confine a gas of ions and electrons, a so-called plasma, inside a doughnut-shaped reactor. More powerful magnets mean less heat loss, allowing more fusion reactions to happen in a smaller, cheaper facility. The change isn't just a little bit: doubling the strength of the magnetic field reduces the volume of plasma needed to produce the same amount of energy by a factor of 16.
Despite billions of dollars of investment in research over the past few decades, no one has yet built a fusion plant that produces more energy than the reactor consumes. But Commonwealth Fusion Systems and its backers are hopeful, and other fusion startups and research efforts have reported recent progress.
Commonwealth Fusion Systems is building a factory to mass-produce magnets and lay the groundwork for a prototype reactor. If all goes well, the start-up plans to provide fusion energy to the grid by the early 2030s.
end password
A single long-term use password will bring crisis to one's own property, which is why security researchers encourage everyone to change passwords frequently, and try to set different passwords for different devices. However, some people will say that if you set too many passwords, it is easy to forget, so what should you do? It is in this situation that biometric technology comes into being.
While biometrics are great for improving security, their transformative benefits don't stop there. By eliminating the need to memorize tedious passwords, biometrics can significantly improve the customer experience. Multimodal biometrics and verification will be adopted as businesses begin to realize the security risks and poor user experience that knowledge-based authentication poses.
Still, it's not that simple. Back in 2014, German hacker Jan "Starbug" Krissler demonstrated how fingerprints could be forged using high-resolution photos of different people's hands, highlighting the potential vulnerabilities of the technology.
Coincidentally, within 24 hours after the release of Apple's iPhone 5s, Stargbug was on the trending list immediately because he successfully "tricked" Apple's TouchID sensor to unlock the phone. It is understood that the fingerprints were extracted through the stains on the screen, and then the phone was unlocked.
Not only fingerprint recognition, but voice recognition also has certain risks. For example, criminals or hackers will consciously record the voice of the victim, and then use this to evade the control of authentication. Perhaps it won't be long before we see a person's face being "reverse-engineered" (an algorithm that only needs some 2D images to complete), and then with the help of a 3D printer, scammers can Walk around wearing the mask of the victim and even withdraw thousands of dollars from ATMs.
In the author's view, not all deficiencies of biometric technologies are discovered by external sources, and similar to other technologies, as the technology is adopted more and more widely, the inherent disadvantages will become clear up. For example, despite improvements in matching accuracy, false positives still plague implementations, which are hard to avoid during the development, configuration, and deployment of technology.
But it is precisely this kind of misjudgment in the police world that urges us to stop and think. Perhaps, the technology's biggest flaw is that the details of biometrics are static. If a password is stolen, there is still an opportunity to change it, but when a database full of information is compromised, an individual's fingerprints, irises or other facial features can no longer be replaced.
So while biometrics is an exciting new technology, its use must be implemented in a dispassionate, planned and strategic manner. For example, when verifying identity, biometric technology must use at least two of them; when helping the police track down criminal suspects, human analysts must be used to confirm the results, and, more importantly, all biometric data must be verified. good storage.
AI protein folding
A few days ago, AlphaFold, a big star in the field of computational biology, made another major breakthrough. It has been able to predict 214 million protein structures from more than 1 million species, covering nearly all known proteins on Earth. The emergence of AI has greatly changed the mode and efficiency of protein prediction. At present, all universities and enterprises have relevant layouts, and related start-up companies in my country will show explosive growth from 2017 to 2021, and most of them have obtained high financing.
Not long ago, Internet giant Meta updated ESMFold, a large protein model. It can predict the complete protein structure directly from the single-sequence language model representation, the accuracy is comparable to AlphaFold, and the inference speed is an order of magnitude faster. Huashen Zhiyao, a domestic AI innovative drug company led by Peng Jian, has also achieved the latest breakthrough: OmegaFold uses a single sequence to determine the 3D structure of a protein. Even if a protein is artificially designed, it can also use AI to predict the 3D structure to determine its function.
1. Computational biology in China will show explosive growth from 2017 to 2021
Computational biology, in essence, is to solve biological problems through computational means. Specifically, it is to build algorithms and models based on different types of biological data (such as concentrations, sequences, images, etc.), so as to understand the biological system itself (such as molecules, cells, tissues, and organs, etc.), and to promote related research and applications. Subject.
From the perspective of application, the current main landing areas include sequence analysis, structure and function analysis, biomolecular dynamics, system modeling, evolution and population genomics, correlation network...
Taking AlphaFold2 as an example, it predicts protein structure based on gene sequence and belongs to the category of structural and functional analysis.
It can be seen that computational biology is an instrumental discipline. To some extent, this determines that there are no strictly computational biology companies on the market, but appear in the name of AI pharmaceuticals, omics, and precision medicine. This is especially evident in our country.
At present, AI pharmaceuticals is the core scenario in China. Not only colleges and universities (Westlake University Institute of Life Sciences, Peking University Frontier Interdisciplinary Research Institute, etc.), Internet giants (Alibaba, Baidu, Huawei, etc.) have relevant research and layout. Related start-up companies have shown explosive growth from 2017 to 2021, and have all received high financing. This situation is also reflected in foreign countries.
According to the report "Investment and Exit Trends in the Healthcare Industry" by Shanghai Pudong Development Silicon Valley Bank, the amount invested in computational biology companies will reach US$5.9 billion (or 39.7 billion yuan) in 2021, an increase of up to three times a year, exceeding the investment in non-computational biology companies. double.
From the perspective of business model, the entire industry is dominated by B2B, mainly for algorithm licensing, biological assets and software usage. In my country, there are mainly the first two types, but in view of the fact that software platforms and pioneering projects can form a closed loop of technology and business iteration.
After the emergence of a large number of advantageous self-developed algorithms, the proportion of software platforms will increase significantly. Foreign countries have begun to commercialize their computational biology services through packaged subscriptions and billing according to usage.
2. The emergence of AI or deep learning has brought a turning point to computational biology
The reason why this track is exploding: firstly, it is related to the explosive growth of deep learning in recent years; secondly, the concept of AI for Science has recently emerged, which makes AI a symbol of landing in the field of biology—computational biology has become a trend . The huge potential brought by the combination of AI and traditional scientific research is expected to bring about a new scientific revolution; finally, for biology itself, it is difficult for traditional experimental and analysis methods to fully develop massive biological data, and computational biology is really needed. Disciplines, while taking into account the comprehensive tools of multiple subdivisions to solve problems.
So, what value can computational biology bring to biology? "Computational Biology In-depth Industry Report" believes that it is divided into two parts: scientific research and application.
The most direct role of computational biology in scientific research is to replace or even surpass experiments. Compared with traditional biological experiments with limited precision in operation level, experimental equipment, observation level, etc., computer-based computational biology not only has lower cost and faster speed, but also theoretically has infinite calculation accuracy and high reproducibility. After internalizing past experience in the AI model, computational biology can propose hypotheses in an automated, large-scale, and parallel manner, so that researchers do not need to rely on a few geniuses, and at the same time lower the threshold for downstream development, which is expected to have a significant impact on the industry landscape. have a major impact. The second is to open up a new method of "assumption first-verification-finally optimize the hypothesis", so that the research and development efficiency can be improved several times.
As early as 1991, Nature pointed out that the starting point of new biological research methods should be that scientists start with theoretical speculations, and then return to experiments to track or verify theoretical hypotheses. Computational biology can just open up a new way of "hypothesis-verification-optimization hypothesis" based on dry-wet cycle experiments, and improve the overall efficiency of biological research and development.
Specifically, on the one hand, the laboratory provides a large amount of available training data for the AI model while quickly verifying the AI prediction through high-throughput wet experiments to improve the accuracy of the AI prediction model. On the other hand, AI will provide hypotheses (high reference value, even practical) that can be verified in wet experiments based on its own data processing capabilities, and the two will jointly iterate and accelerate.
3. Precision medicine will become the long-term focus of computational biology
In the field of AI pharmaceuticals, smart laboratories have become an important manifestation of the company's long-term competitiveness. The value in application can be divided into three categories according to the process:
1. Calculation and deduction of biological properties and principles
Protein structure prediction, pathogenic mechanism research, protein interaction prediction (PPI), epitope prediction of antibodies and antigens, finding the cause of disease or finding new biomarkers based on genomics, etc.
2. Build prediction and judgment models
Target-based compound property prediction in AI pharmaceuticals (mainly involving small molecule drug development), disease diagnosis/monitoring/treatment modeling, biological simulators covering cells/organs/human body, etc.
3. Controlled transformation of organisms
New therapies/drug development, precision medicine and biomanufacturing (represented by synthetic biology). Among them, new therapy/drug development is currently the most mature scenario. Precision medicine will become the long-term focus of computational biology. This is due to the more obvious consumer willingness in the C-end market, the wide use of human bodies, and relatively straightforward product forms.
In this direction, there have been multiple layouts based on multi-omics in foreign countries, while there are relatively few companies related to domestic deployment, and they are all based on genomics, so there is a certain gap.
4. Today's Computational Biology Bottlenecks
It is foreseeable that the future industrial chain of computational biology will be based on data providers as the underlying support + various upper-level related companies (including providing computing platforms and software, molecular modeling/machine learning frameworks, computing power and intelligent experiments) Office of the enterprise) structure.
The "Computational Biology In-depth Industry Report" believes that in order to realize the above expectations, the young computational biology still has the following key bottlenecks to be broken through - some problems are unique to this industry, and some are unique to the entire field of AI science existing:
1) Clarification of the underlying principles of biology
At present, we still have a lot of underlying mechanisms about biology to be thoroughly studied. When carrying out model construction, biological verification, and human landing, we need to introduce this knowledge to reduce deviations that do not conform to domain cognition and ensure accuracy.
2) Unified computing and data framework
Based on microscopic means, some biologically specific problems can be solved, but to be finally implemented, the required model needs to be able to cover multi-omics data, multi-links and functional parallelism. In addition, it is necessary to ensure that various heterogeneous data in computational biology, such as images, videos, molecular maps, DNA codes, gene expressions, electrical signals, etc., have clear standards and common formats for interoperability between different algorithms and platforms. operate.
3) Acquisition of consumer-level data
In the view of analysts, the key industrial development stage of computational biology related to genomics is that data collection has reached consumer-level standards.
4) Project landing ability
At present, there are many machine learning algorithms and models that are quite mature in academics. The key is how to add a specific understanding of biology and make fine adjustments when the underlying data is available. Finally, there is the issue of data privacy, and how to make related models interpretable and gain the trust of this special industry.
Proof of Stake
Cryptocurrencies like Bitcoin use a lot of electricity. In 2021, the Bitcoin network consumed more than 100 terawatt hours, more than Finland’s annual energy consumption.
Proof-of-stake provides a way to build a network that doesn't require a lot of energy. If all goes according to plan, ethereum, the world's second-largest cryptocurrency that runs various applications, will transition to this model in the first half of 2022. This transition is expected to reduce energy use by 99.95%.
Cryptocurrency runs on the blockchain, and the security of the digital ledger generated through transactions must be guaranteed to prevent cheaters, fraudsters, and hackers from invading. Bitcoin and Ethereum currently use a proof-of-work algorithm for security: "miners" solve cryptographic puzzles that compete for the right to validate blocks of new transactions. Successful "miners" are rewarded with cryptocurrency for their work. Proof-of-work means finding solutions to mathematical puzzles, which require a lot of computing power and therefore electricity.
With Proof-of-Stake, validators don't have to fight each other and invest heavily in energy and computing hardware. Instead, their cryptocurrency cache, or stake, allows entry into a sweepstakes. Those chosen gain the power to validate a set of transactions (and thus earn more cryptocurrency). In some networks, validators who exhibit bad behavior are punished by losing a portion of their stake.
Long-term grid energy storage battery
On a sunny afternoon in April 2021, renewable energy broke records on California's main grid, providing enough electricity to meet 94.5 percent of demand. The moment was hailed as a milestone on the road to decarbonisation. But what happens when the sun goes down and the breeze dies down?
Handling the fluctuating electricity production from renewables requires cheap storage for hours or even days, and a new class of iron-based batteries may be up to the task.
Oregon-based ESS, whose batteries can store energy for four to 12 hours, launched its first grid-scale project in 2021. Massachusetts-based Form Energy, which is raising $240 million in 2021, will have batteries that can store electricity for up to 100 hours, and its first installation will be a one-megawatt pilot plant in Minnesota, expected to open in 2021. Completion in 2023.
Both companies opted to use iron-based batteries, one of the most abundant materials on Earth. That means their products could end up being cheaper than other grid storage candidates, such as lithium-ion batteries and vanadium-based flow batteries.
Form Energy says its batteries could end up costing as little as $20 per kilowatt-hour, even less than optimistic projections for lithium-ion batteries in the coming decades.
However, there are still some challenges to be addressed. Iron-based batteries are often very inefficient, which means that a significant portion of the energy put into them cannot be recovered. Plus, side reactions can degrade batteries over time. But if iron-based batteries can be widely deployed at a low enough cost, they could help more people be powered by renewable energy.
AI data generation
Under the background that the epidemic has become the new normal, how to invest in technology has become a concern of business managers. At the end of the year, the market research organization Gartner will publish its forecast on the "important strategic technology trends" in the coming year, one of its most important annual reports, informing corporate management, IT practitioners and government personnel to deal with future investment dynamics and technologies Risk, while guiding technology and investment direction.
Artificial intelligence (AI) plays a significant role in Gartner's technology trend predictions, involving AI Engineering, Hyperautomation, Generative Artificial Intelligence, and Autonomic Systems.
Among them, generative artificial intelligence technology ranks first in Gartner's technology trend forecast and is one of the most eye-catching and innovative artificial intelligence technologies. Gartner predicts that by 2025, generative AI will account for 10% of all data generated, up from less than 1% today.
So-called generative AI, Gartner explained, learns elements from data through various machine learning (ML) methods, and then generates new, completely original, real artifacts (a product or item or task), which are related to the training data. Be similar, not copy.
What are the benefits of generative AI? Gao Ting, senior research director of market research firm Gartner, explained that generative AI can not only judge, but also create. In fact, the biggest use of AI at present is judgment, which means that the use of AI will have structural changes.
"In the past, we let AI keep making judgments and classifications. For example: AlphaGo, you tell me how to move the next move? Call it to make judgments. Or give a photo to an AI model and say: you Help me distinguish whether this is Zhang San, or whether this photo is a photo of a cat." Gao Ting told the interface reporter, "Yes, we will find that in the future, it often needs AI Instead of judging, say, "Come and help me generate a piece of code, what this code does is add from '1' to '100', then AI can automatically generate this code. "
Gao Ting also gave an example that more "synthetic data" can be generated after making a model with existing data. These synthetic data are like human faces. From the perspective of the naked eye, there is no problem with this human face. But in fact, this person is a person who does not exist among the 6 billion people in the world and looks exactly like a real person.
According to Gartner, generative AI learns content or objects from data, and uses data to generate new, completely original new content, enabling the next generation of automatic programming, drug development, visual art, social networking, business services, engineering design and processes. At the same time, it can be used to detect fraud, disinformation and identity theft. But in addition, although technology companies such as Google, Meta, and Microsoft invest the most resources in generative AI, they must also guard against abuses such as deepfakes (Deepfake).
In addition to generative AI, Gartner also pointed out that next year, the trend of AI engineering will also receive industry attention. The engineering of artificial intelligence means that data collection, data processing, modeling, analysis, and report generation are all handled in the form of SOP (standard operating procedure). The seemingly simple work brings great help to data scientists.
"AI engineering is actually not just a technical issue, it is often a procedural issue." Gao Ting said that according to statistics, the most time-consuming work of data scientists when dealing with data engineering is data processing, accounting for 75%. Only 25% of the time is left to define and solve problems, which greatly reduces the ability of enterprises to solve unfamiliar problems. The latest AI engineering can be integrated into industry expertise (Domain Know-How). Gartner believes that by 2025, about 10% of companies that use AI engineering can achieve more than 3 times the rate of return on business.
"In 2020 and 2021, the economy will be affected to varying degrees. Now that the epidemic has become a new normal, many CEOs hope that their business performance will rebound in 2022, or the so-called ' Win back 'the revenue they lost." Gao Ting quoted a Gartner CEO survey report saying that "growth", "digitalization" and "efficiency" will be the three key words of business managers in the coming year. Therefore, the new year's Technological trends are all related to this. In addition to AI technology, new technological trends include privacy-enhanced computing, cloud-native platforms, etc.
"If the main line of technology last year was 'how the world has changed under the influence of the new crown epidemic', this year's is that the new crown epidemic has almost passed or has become a new normal, how to deal with this new normal, Whether it is China or the West, the difference is that everyone handles it differently." Gao Ting said that under the new normal, home office has become the mainstream, and under this circumstance, companies need to use technical means to recover losses under the epidemic, and how to recover in the new situation. Creating a new model under normal conditions to ensure the long-term survival of enterprises has become the main logic of technology narrative in the coming year.
malaria vaccine
RTS,S is the first malaria vaccine approved by the World Health Organization for use in children aged 5 months and older in African areas at medium and high risk of malaria transmission from October 2021.
Malaria is one of the world's three major infectious diseases that seriously endanger human health. With the increasing clinical resistance to various antimalarial drugs such as artemisinin, nearly half of the world's population is still at risk of malaria infection, and the deadliest parasite, Plasmodium falciparum, causes 200 to 300 million infections each year. Since the beginning of the 21st century, about 10 malaria vaccine projects around the world have been approved for clinical trials every year, and about 150 clinical trials have been completed or terminated early.
To date, the malaria vaccine RTS,S is the only vaccine proven to reduce clinical morbidity and mortality in children with malaria. Malaria vaccine RTS,S only had a high protection rate for children aged 5-17 months within one year after 4 doses of vaccination, and then the immune protection rate declined rapidly, and the average protection rate was lower than 30% one and a half years after vaccination. As a breakthrough in the field of malaria vaccine research, the malaria vaccine RTS,S has great practical significance. The World Health Organization expects to save the lives of tens of thousands of African children under the age of 5 every year in the future.
The malaria vaccine RTS,S did not meet the official standards of the World Health Organization, that is, the protection rate is greater than 50%, and the protection time is greater than one year. Therefore, how to effectively curb the prevalence and spread of malaria in tropical and subtropical countries and regions is still an important issue for global malaria research. Scientific problems that need to be solved urgently.
Because the life cycle of Plasmodium includes complex growth stages such as hepatic (cellular) internal stage, red (cellular) internal stage, and mosquito stage, Plasmodium falciparum has highly variable antigenic proteins and variable immune escape strategies, which both limit domestic The development of foreign malaria vaccines is also the main reason why the malaria vaccine RTS,S is not perfect.
In recent years, with the widespread application of a variety of new gene editing technologies in the functional identification of key biomarkers of Plasmodium falciparum, it has become possible for researchers to design multivalent vaccines for different growth stages of Plasmodium falciparum. At the same time, compared with traditional vaccines, the emerging technologies of messenger RNA vaccines, vaccine adjuvants and antigen delivery systems will also provide more potential solutions for malaria vaccine research, making it possible to develop a new generation of highly effective malaria vaccines in the next 5 years. —A key breakthrough will be made within 10 years.
carbon removal plant
Since the Industrial Revolution, human activities have emitted a large amount of greenhouse gases such as carbon dioxide (CO₂), and the greenhouse effect has continued to increase, leading to an increase in the global average temperature.
In fact, even if the world achieves carbon neutrality, since humans have emitted more than one trillion tons of CO₂ since the industrial revolution, it will be a very slow time to reduce the concentration of CO₂ in the atmosphere to the level before the industrial revolution if only relying on natural processes the process of.
As a technology that uses engineering systems to remove CO2 from the atmosphere, the large-scale application of Direct Air Carbon Capture (DAC) technology is of great significance for effectively reducing the concentration of CO2 in the atmosphere and curbing climate change. This technology mainly uses induced draft fan to draw air in, capture CO2 through adsorption, absorption or membrane separation device, and discharge CO2 back to the atmosphere, and the captured CO2 can be stored or utilized. The whole process can be understood as an industrial "photosynthesis effect".
Unlike the CO₂ capture technology for industrial stationary sources, DAC technology can be deployed anywhere in the world with power supply, with more flexible site selection and modular construction.
DAC technology has obvious technical advantages in carbon removal, but the current high operating cost is still a key factor limiting its large-scale application. Recently, researchers from the University of California, Berkeley, looked forward to its development prospects and proposed a policy roadmap suitable for the development of this technology. They believe that the global promotion of DAC technology cannot rely on the market leverage effect, but should promote its large-scale deployment through a continuous "financial incentive + mandatory deployment" policy. From a technical point of view, the key to the development of DAC technology lies in the research and development of efficient and low-cost carbon capture materials and process systems, and its commercial application still needs to rely on technological progress to significantly reduce operating costs.
In recent years, developed countries in Europe and the United States have successively carried out the research and development and application of DAC technology, and continuously reduced operating costs through the progress of materials and technologies. In August 2021, the U.S. Department of Energy announced a $24 million grant to support DAC technology, and some "decarbonization plants" larger than the carbon dioxide capture plant Orca are also under construction. These pioneering works may enable developed countries to master cutting-edge technologies and core intellectual property rights earlier, and seize opportunities for future economic benefits.
COVID-19 Mutation Tracking
The coronavirus is still spreading globally, and of all the nasal swabs that test positive for COVID-19, about two hundred are sent to genetic sequencing machines for additional analysis. The idea is to create a new map of the genome of the SARS-CoV-2 virus and see what changes. This map consists of about 30,000 letters.
Such genetic monitoring allows scientists to quickly detect and alert on new variants such as alpha (a), delta (delta)
and most recently Omicron. It's unprecedented work that makes SARS-CoV-2 the most sequenced organism in history, surpassing influenza, HIV, and even our own human genome. Open databases like GISAID and Nextstrain already display the genetic profiles of more than 7 million pathogens.
Omicron is by far the most mutated variant. In November 2021, a laboratory in South Africa discovered a virus genome with more than 50 mutations in its sequencer, and issued a warning signal for the first time. Almost instantly, computers in Seattle, Boston, and London were using the data to make predictions: Omicron was trouble, a variant that might evade antibodies.
One thing sequencers can't tell us yet is exactly how SARS-CoV-2 will evolve next. That's why some say we should track this virus more closely. Most of the sequences were generated in places such as the UK, US and Denmark, but in areas without sequencing capacity, viruses can still evolve unknowingly. Fortunately, South Africa's rapid work in finding Omicron and tracking its spread provided early warning to the world.
Blue Ocean Brain Life Science Solutions
1. Case overview
A school of pharmacy started in 2021. The research and development of drug molecules is a very complicated and time-consuming process, and drug molecular screening is only a part of the preliminary process. For example, looking for small molecules that bind to proteovirus enzymes, due to the existence of ligand (small molecule) libraries of different types or research institutions, the number of ligand (small molecule) libraries is huge, and the number of ligands in each ligand library is tens of thousands (or even larger), it is impractical to test the verification experimentally. Screening through computer numerical simulation, scoring the binding effects of different ligands, screening out some ligands with high scores and reasonable binding modes as candidate drugs for experimental verification, can effectively accelerate the drug development process.
Due to the huge number of ligand libraries, completing the screening within a limited time is also a huge challenge. For example, the ligand library has 10,000 candidate ligands, and the average processing time for each ligand is 1.5 hours, requiring a total of 15,000 hours (625 days). Therefore, in order to complete the calculation within the specified time, the following conditions are required:
1) A computing platform with powerful computing power
2) Large-capacity storage for storing processing data and calculation results
In addition, in order to ensure that the screening calculation can be completed efficiently and smoothly, calculation services are also required, including:
1) The cluster software operating environment ensures that the software runs in a multi-machine environment and data access
2) A parallel solution that can support concurrent processing of multiple tasks in a multi-machine environment
In addition to computing platforms, drug screening requires high-performance application software. Drug screening simulation calculations include Docking and molecular dynamics calculations: Docking is relatively time-consuming and is often used for preliminary screening of a large number of ligands. The main software includes dock6, Autodock Vina, Glide, etc. The molecular dynamics simulation calculation is time-consuming, and the time change of the test effect is used to further analyze the results of the primary selection of Docking. The main software includes Gromacs, Namd, Amber, etc. The effect of using GPU acceleration is generally more obvious.
2. Program and value
The research and development of small drug molecules requires high-performance clusters with powerful computing power. Obtaining these computing resources and services has become the top priority at the moment. Tsinghua University School of Pharmacy has built four A100 liquid-cooled servers, nine CPU servers, and two high-throughput liquid-cooled servers to provide a basic computing platform for high-performance computing environments.
When using DOCK6 to deal with the docking case of the ligand (small molecule) library, a large number of small molecule files are stored in a folder, such as mol2. Viral protease) for calculations. At this time, it is necessary to provide a large number of GPU and supercomputing products in a short period of time, as well as round-the-clock technical support. Build an open shared platform and use high-performance computing clusters for molecular docking, molecular dynamics simulation, and deep learning model training for drug development, shortening the computing work that takes days to hours, and increasing the speed by 8 to 20 times. At the same time, different sub-accounts are created for each R&D teacher to realize computing resource sharing and data sharing. To provide a basic computing platform for a high-performance computing environment, high-throughput task solutions are also required to achieve efficient drug screening.
3. Summary
The drug research and development of the School of Pharmacy requires high-performance computing clusters with powerful computing power. For example, drug screening requires Docking processing of a large number of small molecules. Teachers of the School of Pharmacy can use Blue Ocean Brain's high-performance liquid-cooled servers to quickly build high-performance clusters and obtain high-performance computing instances to meet computing power needs. At the same time, it provides a solution for high-throughput task processing, enabling drug screening to be processed concurrently on multiple computing nodes and multi-cores, reducing the overall task execution time.