AlphaFold Changed Science. After 5 Years, It’s Still Evolving
SMRTR summary
Five years ago, Google DeepMind shifted from teaching AI to master the ancient game of Go to tackling one of biology's greatest mysteries: how proteins fold into their three-dimensional shapes. The result was AlphaFold, which has since compiled a database of over 200 million protein structures used by 3.5 million researchers worldwide and earned a Nobel Prize in Chemistry last year.
Pushmeet Kohli, who leads DeepMind's AI for Science division, describes their approach as solving "root node problems" where breakthroughs unlock entire branches of research. The latest iteration, AlphaFold 3, now predicts interactions between proteins, DNA, RNA, and drugs, though it faces challenges with "structural hallucinations" in disordered protein regions.
DeepMind is now developing an "AI co-scientist" built on Gemini 2.0 that generates and debates hypotheses. Researchers at Imperial College recently used it to study viruses that hijack bacteria, potentially opening new paths to combat antimicrobial resistance. Kohli envisions AI compressing the hypothesis generation phase while humans focus on designing experiments and interpreting significance.
The ultimate goal remains audacious: simulating an entire human cell to revolutionize drug development and personalized medicine.
SMRTR provides this summary for quick context. The original article belongs to Wired.
Read the original article