#427: 10x Genomics Is Enabling Breakthroughs In AI-Driven Healthcare, & More

Recent breakthroughs in single-cell genomics and AI are transforming healthcare, unlocking knowledge that should advance drug discovery and personalized medicine. Single-cell genomics helps researchers analyze the genetic activity within individual cells to understand complex diseases like cancer and identify which cells are implicated in disease progression.

AI is critical¹ to organizing and interpreting the vast datasets generated by single-cell genomics. Foundation models like scGPT² and scFoundation³ are trained on millions of cells, enabling not only the classification of cell types but also the prediction of cellular responses to drugs and the integration of complex tissue datasets. The models should help accelerate drug discovery by predicting outcomes in silico, dramatically reducing the time and cost of developing new therapies. The technologies offered by companies like 10x Genomics are playing an important role by generating vast amounts of high-quality single-cell data to enable AI breakthroughs.

Ultimately, the convergence of AI and single-cell genomics could lead to the creation of virtual cells, enabling researchers to simulate cellular behavior in different disease and treatment regimes. Helping to develop new therapies faster, safer, and more efficiently—without involving labs—virtual cells could revolutionize drug discovery.

Last week, BitGo, an institutional custodian for digital assets, announced its intention to expand internationally and enhance the security of WBTC—Wrapped Bitcoin—through a joint venture with Bit Global. The initiative would make WBTC the first multi-jurisdictional and multi-institutional custodial product.

WBTC is the most popular solution for bridging bitcoin onto the Ethereum chain. While holding custody of the native bitcoins, BitGo issues an ERC-20 token representing the underlying bitcoin in a format compatible to the Ethereum blockchain.

The digital asset community reacted cautiously to this announcement because of Bit Global's close association with Justin Sun, the controversial founder of the Tron blockchain who has been involved in a number of questionable projects.⁴ BitGo’s venture with Bit Global—and by extension, Justin Sun—would give each company control over one of the two keys in the multi-signature setup securing WBTC.

WBTC is a crucial collateral asset in many DeFi protocols: more than 150,000 WBTC representing approximately $9 billion are on the Ethereum blockchain, as shown below. The BitGo/Bit Global venture could be particularly problematic for MakerDAO and Aave, which hold $400 million and $2.2 billion, respectively, of WBTC.

Source: Dune Analytics. 2024. “WBTC,” as of August 16, 2024. For informational purposes only and should not be considered investment advice or a recommendation to buy, hold or sell any security or cryptocurrency.

Given the risks associated with the BitGo/Bit Global partnership and its multi-signature setup, both MakerDAO and Aave have recoiled with DAO (Distributed Autonomous Organizations) proposals to offboard WBTC as a collateral asset. Meanwhile, Coinbase shrewdly has announced the launch of cbBTC on Base,⁵ although details have yet to be disclosed. Managed by Coinbase, cbBTC is likely to compete directly with WBTC.

Sakana AI, a Tokyo-based AI research lab founded by former Google employees, recently released details on a new project called “The AI Scientist.” This “agentic”⁶ system is designed to automate the entire scientific process by combining the powerful capabilities of large language models, including language comprehension, planning, reasoning, and coding skills. AI Scientist follows the process illustrated below.

Source: Sakana AI. For illustrative purposes only.

Given a starting template that directs the research—for example, exploring methods to lower the cost of training AI systems—“AI Scientist” will brainstorm ideas, check their novelty, design and execute a series of experiments, and then write an analysis ready for peer review. The Sakana team even designed a system to judge the quality of the papers created by AI Scientist, discerning whether the papers it produced could meet the criteria for admission to a machine learning conference, though output quality can vary, and we do not know how the papers would stand up to a true peer-review process. Impressively, each paper costs only $15 to generate.

That said, the project is far from perfect or complete. The system cannot interpret images, for example, and makes simple mistakes when comparing numbers. Importantly, while the product does seem able to advance existing research, its capacity to generate new research ideas is unclear.

Long a dream in the AI research community, AI systems seem to be reaching a tipping point in the automation of scientific discovery. Former Open AI employee Leopold Aschenbrenner believes that automating AI research is key to unlocking superintelligence capabilities: once we automate the scientific process, so his theory goes, scientific breakthroughs will depend on little more than the budget for computing power.