AlphaFold

The story of AlphaFold

Google DeepMind

NotebookLM shared with system prompt and other contexts

Click Crash Courses for grounding sources in NotebookLM

AlphaFold is an artificial intelligence program developed by Google DeepMind that predicts a protein’s 3D structure from its amino acid sequence. It essentially solved the 50-year-old “protein folding problem”, achieving an accuracy level that competes with traditional, expensive experimental methods like X-ray crystallography. [1, 2, 3, 4, 5]

Because of its massive impact on science, its co-creators Demis Hassabis and John Jumper were co-awarded the Nobel Prize in Chemistry. [6]

🧬 Evolution of the AlphaFold Models

  • AlphaFold 1 (2018): Showed initial proof-of-concept success at the CASP13 competition by predicting relationships between amino acid pairs. [7]
  • AlphaFold 2 (2020): Redesigned using an attention-based neural network architecture called the Evoformer. This model achieved near-experimental accuracy and predicted the structures of nearly all catalogued proteins. [1, 5, 6, 7, 8, 9]
  • AlphaFold 3 (2024): Expanded capabilities beyond proteins. It maps interactions between DNA, RNA, chemical modifications, and drug ligands. [6, 10, 11]

💻 Core Technology and Process

AlphaFold processes genetic text strings into structural geometry through three main steps: [5, 9]

  1. Multiple Sequence Alignment (MSA): The AI searches genetic databases to align similar proteins across different species, identifying co-evolving amino acids.
  2. Evoformer Network: A deep learning module that processes evolutionary history and spatial constraints concurrently.
  3. Structure Module: Refines the final data by rotating and translating individual amino acids into a physical 3D structure. [5, 7, 9, 12]

🌍 Real-World Impact and Access

  • AlphaFold DB: Created in partnership with EMBL-EBI, this database provides open, free access to over 200 million protein structures to the global research community. [1]
  • AlphaFold Server: A free platform where scientists can run non-commercial biomolecular predictions with a few clicks using AlphaFold Server. [10, 11]
  • Applications: Used globally by millions of researchers to accelerate drug design, combat antibiotic resistance, model heart disease factors, and engineer climate-resilient crops. [5, 6, 13, 14]

If you are a researcher looking to utilize AlphaFold, would you like to know how to navigate the AlphaFold Protein Structure Database, or are you looking to run your own prediction sequences on the AlphaFold Server?

[1] https://alphafold.ebi.ac.uk

[2] https://en.wikipedia.org

[3] https://www.youtube.com

[4] https://deepmind.google

[5] https://www.youtube.com

[6] https://deepmind.google

[7] https://www.youtube.com

[8] https://www.embl.org

[9] https://www.youtube.com

[10] https://alphafoldserver.com

[11] https://blog.google

[12] https://www.ebi.ac.uk

[13] https://pmc.ncbi.nlm.nih.gov

[14] https://deepmind.google

Top 5 Crash Courses

AlphaFold

Sutanu M

AlphaFold Decoded – Implementing AlphaFold 2 from Scratch

Kilian Mandon

AlphaFold

Yin-Phin Lim. 颖斌林。

AlphaFold: Protein Structure Prediction

DevStoriesEU

AlphaFold

Márton Vaitkus