Recently, Insilico Medicine, an artificial intelligence-driven drug development company, has made breakthroughs one after another.
On February 17, Silicon Intelligent nominated a preclinical candidate compound that targets glutamine peptide cyclotransferase-like (QPCTL) for innovative tumor immunotherapy. This is also the first milestone reached by The strategy of Silicon Intelligent and Fosun Pharma, which is jointly conducting research on the project for new drug clinical trial applications.
A week later, Silicon Intelligence announced again today (February 25) that the first drug candidate ISM001-055, a new target with a new chemical structure discovered by artificial intelligence, has officially entered Phase I clinical trials.
It is reported that the Phase I clinical trial was designed with double-blind, placebo-controlled, single- and multiple-dose increments to evaluate the pharmacokinetic profile, tolerability, and safety of drug candidates.
Compared with these major milestones, Silicon Intelligence is more concerned by the foreign AI pharmaceutical circle, but it is a paper recently published on the preprint website ArXiv. The paper describes a case of linking AlphaFold with its self-developed artificial intelligence platform to quickly identify potentially potentially therapeutic hepatocellular carcinoma emerging compounds within 30 days.
图丨AlphaFold Accelerates AI Powered Drug Discovery：Efficient Discovery of a Novel CDK20 Small Molecule Inhibitor（来源：ArXiv）
How does Alphafold land in the actual scenario of drug development? DeepTech started with this paper and engaged in an in-depth conversation with Dr. Alex Zhavoronkov, founder and CEO of Silicon Intelligence.
Figure | Dr. Alex Zhavoronkov, Founder and CEO of Silicon Intelligence (Source: Silicon Intelligence)
Challenge the limits of the speed of discovering emerging compounds
According to reports, in the past, traditional drug research and development took an average of one year from target identification to emerging compounds. Today, many other methods, such as DNA-coding libraries and more advanced virtual screening, can help us achieve this process as quickly as possible.
In a Nature review paper, "How to Improve R&D Efficiency: A Major Challenge for the Pharmaceutical Industry," it is also mentioned that emerging compounds can now be found on average in about half a year after a new target is launched.
However, for an AI pharmaceutical company with a mission to accelerate the development of new drugs, Alex and his team challenged this time.
They wanted to show that the linkage of AI platforms could further shorten the process from disease target identification to discovering emerging compounds, and AlphaFold will be an important part of achieving this challenge. "AlphaFold has been known for a long time, but when it is combined with end-to-end artificial intelligence, AlphaFold's role will play faster and better." Alex said.
In fact, this isn't the first time That Silicon Intelligence has challenged the limits of speed in drug discovery. In 2019, the company used AI technology to design and screen a DDR1 receptor inhibitor in just 46 days, setting a new record for the fastest discovery of a drug-lead compound.
The compound has been tested to have good pharmacokinetic properties in mice and is expected to continue to optimize and potentially be used in the treatment of fibrosis-related diseases. The findings were also published in Nature Biotechnology in 2019.
图| Deep learning enables rapid identification of potent DDR1 kinase inhibitors（来源：Nature Biotechnology）
AlphaFold is linked to an end-to-end AI platform
"This is the first time a research team has combined end-to-end AI with AlphaFold." Alex said.
Founded in 2014, Silicon Intelligence has developed an end-to-end artificial intelligence platform Pharma.AI using machine learning technologies such as deep learning, generative adversarial networks, and pre-trained models. PandaOmics and Chemistry42, the target recognition engine and small molecule generation engine that linked with AlphaFold to discover emerging compounds, are an important part of the AI platform.
Figure丨 Silicon Intelligent Artificial Intelligence R&D Platform Pharma.AI (Source: Silicon Intelligence)
In the process of traditional drug development, pharmaceutical companies will first select targets from the literature, or communicate with scholars and doctors to understand potential targets. In this case, PandaOmics took on this part. "PandaOmics is a target discovery engine, just like Google's search engine. You can enter the name of the disease and view the data information already in the system. Alex said.
Through PandaOmics' search, the research team targeted 20 potentially relevant targets for hepatocellular carcinoma. "Among them, we chose a target cdK20 that has no structure, but alphaFold has predicted its crystal structure."
Alex further explained that previously, researchers could not design and generate small molecule compounds through virtual screening for targets like CDK20 that do not have a crystal structure. The only method – high-throughput screening is not only time-consuming, but also expensive.
Then we moved on to the chemistry42 work, which designed compounds to match the CDK20 target crystal structure predicted by Alphafold. Subsequently, the R&D team rapidly synthesized and tested these compounds, and found 1 active emerging compound in 7 compounds.
The entire process from target identification to finding the emerging compound took only 30 days. It is reported that for the first time, a research team has combined end-to-end AI with AlphaFold.
A stone stirred up thousands of waves, causing the attention of the AI pharmaceutical circle
The R&D team wrote the whole process into a paper to share on the preprint website, which triggered a discussion in the AI pharmaceutical circle. "I shared this paper on social platforms and we quickly received some feedback, such as the incipient compound is not very active, only 8 micromoles, or the acquisition of the seedling compound is only accidental." Alex said.
In the face of industry comments, the R&D team further optimized the seedling compound, Alex introduced, "We used a ligand-based drug design to optimize the original activity of 8 micromoles of the seedling compound. Now the growth compound activity has reached 200-300 nanomoles, which is quite impressive. "Counting the Chinese New Year holiday, the whole process added another 20 days.
"We've updated the paper on ArXiv and I've re-shared it on social networks. We wanted to get as much feedback as possible from the outside world so that this attempt would be a classic example of applying an AlphaFold database. Interestingly, this achievement also received the attention and response of Deepmind founder Demis Hassabis, who said in an email, "I am very pleased to hear that AlphaFlod has helped accelerate this work." ”
Asked about the value alphafold will also extract value in those scenarios, Alex said, "In many other areas, AlphaFold has a greater impact, such as studying and predicting protein-to-protein interactions, dynamic protein folding, discovering additional docking pockets, understanding the effect of mutations on structure, and many other applications." ”
Incorporate end-to-end AI into the drug discovery process
So, is the case of a fledgling compound found within 30 days replicatable? Alex said that artificial intelligence needs to be closely linked to experimental teams and infrastructure, and there are many excellent artificial intelligence companies in the market, but without a very professional and talented global drug discovery team, it is impossible to develop drugs quickly and effectively.
"Although I came up with the original idea myself, almost all of the work involved in the paper was done by China's core drug development team. Dr. Ren Feng is a true drug discovery expert who used more traditional methods to discover drugs, and now he has successfully incorporated artificial intelligence into the company's early drug development and quickly developed and executed plans to bring this concept to the ground. ”
Figure | In June 2021, The group photo of the night of the release of the Series C financing by InsiliCo, from left to right: Dr. Ren Feng, Chief Scientific Officer of Insili Intelligence, Dr. Kai-Fu Lee, Chairman and CEO of Innovation Workshop, Dr. Alex Zhavoronkov, Founder and CEO of Insili Intelligence, and Yang Xiaolong, Partner of Innovation Workshop (Source: Innovation Factory)
Alex believes that "integrating end-to-end AI into the drug development process is the future of biopharmaceuticals." "The crystal structure predicted by AlphaFold alone is just the starting point, and then industry-leading AI systems are needed to quickly generate promising molecules.
Empowered by end-to-end artificial intelligence, Insil intelligent has nominated five preclinical candidate compounds in the past 12 months, including two drug candidates for the treatment of fibrosis, two PHD2 small molecule inhibitors that target the Nobel Prize pathway, and small molecule inhibitors for cancer immunotherapy in collaboration with Fosun Pharma.
-End-
reference:
1.Steven M. Paulet.al, Nature Reviews Drug Discovery9, 203–214 (2010)https://www.nature.com/articles/nrd3078