AI-Generated Drug Enters Phase II Clinical Trials - H+ Weekly - Issue #421
This week - DeepMind promises to beat OpenAI; an AI startup with no product raises €150M; industrial-scale stem cells are here; self-improving robotic agent from DeepMind; and more!
When DeepMind released AlphaFold 2 in 2020, many people speculated about the impact that AI would have on medicine. AlphaFold successfully solved the protein folding problem, one of the biggest challenges in computational biology, and opened up new possibilities in medical research.
This week, Insilico Medicine announced that a drug discovered by artificial intelligence has entered Phase II clinical trials in the US and China.
Insilico Medicine was founded in 2014 by Alex Zhavoronkov as an alternative to animal testing for research and development programs in the pharmaceutical industry. They utilize AI and deep-learning techniques to analyze how a compound will affect cells and identify potential drugs for treating cells, as well as predict possible side effects. To date, the company has raised $400 million to realize this vision.
Their main product, Pharma.ai, is an AI-powered end-to-end drug discovery platform that can identify and prepare potential molecules for use in disease treatment.
Under the hood, Pharma.ai uses three AI systems - DeepMind's AlphaFold, as well as in-house systems PandaOmics and Chemistry42.
First, PandaOmics sifts through research papers and omics data to identify possible therapeutic targets that may have been overlooked in the past, with ample evidence to support the connection.
Next, AlphaFold predicts the structure of the protein causing the problem, and Chemistry42 generates molecules based on the predicted structure.
The most promising candidate molecules are then synthesized and tested in the laboratory on real cells.
According to the paper published by Insilico, this entire process took only 30 days to identify and validate seven candidates for a new treatment. To put this speed into context, it typically takes screening over a million molecules to approve just one new treatment. The journey from discovery to approval usually spans 12 to 15 years and requires an investment of around $1 billion. Insilico promises to significantly accelerate this process.
This week, Insilico Medicine initiated Phase II clinical trials for a drug discovered and designed by AI. The drug, called INS018_055, is aimed at treating idiopathic pulmonary fibrosis (IPF), a disease that affects five million people worldwide. Once diagnosed, individuals with IPF typically have a life expectancy of only three to four years. Insilico is currently recruiting 60 patients with IPF in China and the US, who will undergo a 12-week dosage of the drug. In early 2023, INS018_055 produced positive topline data in Phase I trials conducted in New Zealand and China.
While Insilico Medicine may be the first company to initiate Phase II clinical trials using AI-generated drugs, it is not the only one operating in this field. Shortly after the release of AlphaFold 2, DeepMind spun off Isomorphic Labs with the aim of "redefining drug discovery with the power of artificial intelligence."
Companies applying AI to drug discovery, such as Insilico and Isomorphic Labs, have the potential to revolutionize medicine by finding treatments for various diseases, including rare genetic conditions, and cancer, and even address ageing.
Insilico itself is preparing for an IPO in Hong Kong (the company is based in Hong Kong and New York). Forbes reports that the company has not disclosed specific details of the IPO in its filing to the Hong Kong stock exchange, but there are reports suggesting that Insilico plans to raise $200 million. In November 2021, Bloomberg News reported that the company had previously filed confidentially for a U.S. IPO with the intention of raising approximately $300 million.
From H+ Weekly
Becoming a paid subscriber now would be the best way to support the newsletter.
If you enjoy and find value in what I write about, feel free to hit the like button and share your thoughts in the comments. Share the newsletter with someone who will enjoy it, too. That will help the newsletter grow and reach more people.
🦾 More than a human
In this talk at TED Conference, neurotechnologist and cofounder of OpenBCI Conor Russomanno shares his work building brain-computer interfaces that could enable us to control the external world with our minds. He discusses the quickly advancing possibilities of this field - including the promise of a "closed-loop system" that could both record and stimulate brain activity - and invites neurohacker Christian Bayerlein onto the stage to fly a mind-controlled drone by using a biosensing headset.
Police got called to an overcrowded presentation on “rejuvenation” technology
So many people wanted to attend Juan Carlos Izpisua Belmonte’s lecture at The International Society for Stem Cell Research Conference in Boston on using stem cells in anti-ageing research that the organisers had to call the police to restore order. “It’s a good sign for this field that there is so much interest. It’s a hot topic. Hotter than we expected,” said Haifan Lin, a Yale University cell biologist and president of the International Society for Stem Research, which organized the meeting. Juan Carlos Izpisua Belmonte is a Spanish scientist known for his work on rejuvenation through cell reprogramming who is now working at Altos Labs - a secretive, wealthy anti-ageing startup.
The wild race to improve synthetic embryos
Two weeks ago, The Guardian published an article discussing a breakthrough in the development of synthetic human embryos (I reported on that in Issue #419). This sparked controversy within the scientific community. Antonio Regalado argues that The Guardian's article is misleading, as it implies that the race to create synthetic human embryos had been successfully concluded, despite the lack of scientific evidence supporting the results cited. Interestingly, there is a twist to this story. Shortly after The Guardian's publication, Jacob Hanna, an Israeli scientist, released a preprint paper detailing the creation of highly realistic synthetic embryo models that had been cultivated for approximately 14 days. This whole controversy shows how competitive the race to create synthetic human embryos is.
🧠 Artificial Intelligence
Google DeepMind’s CEO Says Its Next Algorithm Will Eclipse ChatGPT
Wired reveals more details about Gemini - Google’s secretive next-generation large language model, first revealed at Google I/O 2023. So far we knew Gemini will be multimodal and capable of planning actions. Now we’ve learned that DeepMind will add its deep experience with reinforcement learning to give Gemini novel capabilities. “At a high level you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models,” Demis Hassabis, CEO of Google DeepMind says. “We also have some new innovations that are going to be pretty interesting.”
Inflection-1: Pi’s Best-in-Class LLM
Infection AI announces the release of Inflection-1 - their in-house large language model. According to the company, Infelction-1 is the “ best model in its compute class, outperforming GPT-3.5, LLaMA, Chinchilla, and PaLM-540B on a wide range of benchmarks commonly used for comparing LLMs”. If you want to learn more about Inflection AI, check my post for AI Supremacy where I profiled them and five other companies in the generative AI space.
France's Mistral AI raises 105 million euros shortly after being set up
The AI hype is real. French AI startup Mistral has raised 105 million euros ($113.4 million) - Europe’s largest-ever seed round - just a month after being set up. The company has no products and barely any employees. What the company has, however, are founders with good CVs who previously worked at DeepMind and Meta. Two of them were behind the team that developed Meta’s open-source Llama language model.
The people paid to train AI are outsourcing their work… to AI
A significant proportion of people who were hired to help train AIs (for example, by labelling the training data) have been using AI to do the job, a new study has found. The study estimated that somewhere between 33% and 46% of the workers had used AI models like OpenAI’s ChatGPT. The problem this might cause is that human input is still necessary in the training process to spot errors and set the algorithm on the right track. Training these systems on outputs generated by other AI systems can amplify these errors and reduce the accuracy of AI models over time.
DeepMind has put their multimodal model Gato (Spanish for “cat”) into a robot and created RoboCat - a state-of-the-art model that can quickly learn new skills. RoboCat successfully learned how to operate new robotic arms and solve complex tasks with as few as 100 demonstrations in their experiments. The paper explaining RoboCat in detail is available on ArXiv.
A Wearable Robotic Assistant That’s All Over You
Meet Calico - a tiny robot that crawls over you on a special track sewn into the garment. Researchers envision their creation can be used to sense vital signs, listen to organs or detect falls. It could also be used as a coaching assistant or just be an adorable and entertaining companion. BTW this article was written by Mat Sadowski who runs the Weekly Robotics newsletter, an excellent source of the latest developments in robotics.
While some researchers look to nature for inspiration, a team from Japan has chosen to use nature. In their paper “Biological Organisms as End Effectors”, they demonstrate that pill bugs and chitons—types of small, harmless creatures—can be utilised as functional grippers. This reminds me of that research when engineers used dead spiders as robotic manipulators. The difference is that the researchers from Japan did not harm the insects and released them after the experiments.
Amazon’s New Robots Are Rolling Out an Automation Revolution
Thanks to advancements in AI and robotics, a new wave of robots is coming to Amazon’s warehouses to further improve the efficiency of the e-commerce giant. Equipped with 3d scanners to better understand their surroundings and expressive displays to communicate with humans, robots, such as Proteus and Sparrow, are designed to work alongside humans and perform tasks previously done by people.
Industrial-Scale Stem Cells Are Here
At the International Society for Stem Cell Research, Bit.bio announced a new technique for producing stem cells with an unprecedented level of consistency. The company claims that its approach can significantly improve the consistency, scalability, and cost-effectiveness of cell manufacturing. Having a reliable source of high-quality stem cells will impact regenerative medicine research and could enhance CAR T-cell therapy for cancer by providing more consistent cell preparations. Other applications include screening new drugs and maybe even 3d printing organs.
H+ Weekly sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.
H+ Weekly is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
A big thank you to my paid subscribers and to my Patreons: whmr, Florian, dux, Eric and Andrew. Thank you for the support!
Thank you for reading and see you next Friday!