RT Pushmeet Kohli (@pushmeet)At @GoogleDeepMind, we believe AI is the ultimate catalyst for science. 🧬The best example of this has been the AlphaFold database (AFDB) of protein structure predictions which has been used free of cost by more than 3.3 millions researchers across the world!Today, in collaboration with @emblebi, @Nvidia and @SeoulNatlUni, we are expanding the database by adding millions of AI-predicted protein complex structures to the AlphaFold Database. To maximise global health impact, we’ve ...

AI for Science & AlphaFold Overview - AI systems can accelerate scientific discovery and enable new breakthroughs, particularly in healthcare [1] - AlphaFold, a system developed for protein structure prediction, has been cited approximately 35,000 times, demonstrating its impact on scientific research [1] - The speaker's guiding principle is to build tools that enable scientists to make discoveries [1] Protein Structure Prediction & Biological Significance - Proteins, numbering around 20,000 different types in humans, perform nearly every function in a cell [1] - Determining protein structure is exceptionally difficult, often requiring years of effort and significant resources, costing around $100,000 [2] - There are approximately 200,000 known protein structures, with roughly 12,000 new structures being added annually [2] - Protein sequence discovery is happening approximately 3,000 times faster than protein structure determination [2] AlphaFold Development & Key Factors - AlphaFold's success was driven by data (200,000 protein structures), compute (128 TPU V3 cores for two weeks), and, most importantly, research and innovative ideas [2] - Research and novel ideas were approximately 100 times more valuable than the data used in training AlphaFold [3] - Mid-scale ideas, rather than just scaling transformers, are crucial for building transformative AI systems [2][3] Impact & Applications of AlphaFold - AlphaFold has enabled scientists to make discoveries in areas like vaccine and drug development, and understanding how the body works [1] - The release of the AlphaFold database, containing approximately 200 million protein structure predictions, significantly increased its adoption and impact [3] - Researchers are using AlphaFold in unexpected ways, such as predicting protein interactions and engineering proteins for targeted drug delivery [5][6] - AlphaFold is estimated to have accelerated the field of structural biology by approximately 5-10% [9]