Google, Meta to Build Biological Data for AI Training [Global AI Briefing]

[Financial News] Google DeepMind, Meta Platforms and Isomorphic Labs, a Google-affiliated AI drug development company, are setting out to build large-scale biological data for AI training.
AI can be used effectively in the biotech industry for drug development and disease treatment, but the shortage of vast amounts of data for AI to learn from has been a bottleneck. Google and Meta have now decided to generate the data themselves.
According to a report on the 7th (local time) by IT publication The Verge, Google, Meta Platforms and Isomorphic Labs have agreed to invest $300 million (about 400 billion won) in the Virtual Biology Initiative, a project of Biohub, a nonprofit research organization founded by Meta CEO Mark Zuckerberg and Priscilla Chan, to build biological data for AI training.

The goal of the Virtual Biology Initiative is to build large-scale biological data that will enable AI to understand and predict how actual cells behave.
The project will compile previously accumulated genetic information and research findings while creating “virtual cells” for use in disease research, allowing researchers to first simulate cellular responses on computers instead of testing every possibility one by one.
The goal is to create AI models that can predict cellular responses and test various treatments, substantially reducing the time and cost involved in finding drug candidates.
The lack of vast amounts of data for generative AI such as ChatGPT to learn from has long been cited as an obstacle to the use of AI in the life sciences.
Drug development takes a long time because it involves repeatedly conducting numerous experiments, administering candidate compounds to actual cells or animals. Using AI in experiments could reduce the time they take, but there was not enough data measured consistently on how actual cells respond to specific drugs or changes in their environment. As a result, even the most advanced AI models could not learn the biotech field, making it difficult to use them directly in drug development.
Google and Meta plan to address this bottleneck in the biotech industry by establishing a system that uses virtual-cell experiments to generate large amounts of new data and then uses that data to train AI, enabling AI to be applied to drug development.
Biohub said, “We aim to compress data-building work that would normally be expected to take several decades into five years, produce the first dataset in about a year, and build an accurate predictive model within five years,” and expressed hope that drug development and biological research would be significantly accelerated.
[email protected] Lee Gu-sun Reporter