Loading
What changed between revision 1 and revision 2, compared over the article's markdown source with its frontmatter — so a change to status, evidence or reversibility shows up as the edit it is.
Older · revision 1
Import· initial corpus import, not an edit
Initial import of content/ai-protein-design.md — the filesystem corpus, unchanged. Not an edit.
17.1 kB
Revision 1
Revision 2
lines 33–39 → 33–3932 unchanged lines not shown
**AI protein design** covers two distinct uses of machine learning on proteins: predicting the**AI protein design** covers two distinct uses of machine learning on proteins: predicting thethree-dimensional structure a given amino acid sequence folds into, and generating new sequencesthree-dimensional structure a given amino acid sequence folds into, and generating new sequencesthat fold into a shape chosen in advance. The first was largely solved for single chains betweenthat fold into a shape chosen in advance. The first was largely solved for single chains between2018 and 2021, mostly by DeepMind's AlphaFold. The second is being solved unevenly, by generative2018 and 2021, mostly by DeepMind's [AlphaFold](https://aiwiki.ai/wiki/alphafold). The second is being solved unevenly, by generativemodels from David Baker's laboratory and others, and it is the half that produces molecules whichmodels from David Baker's laboratory and others, and it is the half that produces molecules whichhave never existed. AlphaFold predicts; it does not design.have never existed. AlphaFold predicts; it does not design. lines 78–84 → 78–8438 unchanged lines not shown
alone drop the alignment, trading some accuracy for speed and for coverage of sequences with noalone drop the alignment, trading some accuracy for speed and for coverage of sequences with noknown relatives.known relatives. Design runs in two stages. A generative model such as RFdiffusion, adapted from a structureDesign runs in two stages. A generative model such as [RFdiffusion](https://aiwiki.ai/wiki/rfdiffusion), adapted from a structurepredictor by training it to denoise corrupted backbones, produces a protein shape to order: a binderpredictor by training it to denoise corrupted backbones, produces a protein shape to order: a binderagainst a specified epitope, a symmetric assembly, or a scaffold holding catalytic residues inagainst a specified epitope, a symmetric assembly, or a scaffold holding catalytic residues inposition.[^watson2023] A second model, ProteinMPNN, then reads thatposition.[^watson2023] A second model, ProteinMPNN, then reads thatremoved, struck through added, underlinedLine numbers count the serialised markdown of each revision, frontmatter included.
236 lines are the same in both revisions, and 224 of them are not shown. Neither revision here is necessarily the current article — read AI protein design for the text this wiki stands behind today.