Annotated Papers by Chris Hayduk

By Chris Hayduk

Biological AI
papers, annotated.

Detailed notes on research papers, with code from my PyTorch implementations. I explain the notation, show the tensor operations, and use interactive figures where a static diagram falls short.

Read AlphaFold2 About the annotations

Free to read, with no account required.

Protein structure prediction AlphaFold2
01 / INPUTAmino acid sequence
02 / MODEL

Evoformer Structure Module

Sequence and pair representations become 3D coordinates.
03 / STRUCTURE

Protein
backbone

The example structure
used in Chapter 14
A simplified view of the AlphaFold2 architecture.
Topics covered
  • Protein structure
  • Attention
  • Evolution
  • 3D geometry
  • PyTorch

Papers

Available annotations

1 paper

Featured annotationNature · 2021

The Annotated AlphaFold2

A detailed explanation of AlphaFold2’s architecture, from the input features to the Evoformer and Structure Module. Code excerpts come from minAlphaFold2, a compact PyTorch implementation.

Based on John Jumper et al. · Annotated by Chris Hayduk

Chapters
16
Code excerpts
37
Interactive figures
8
Start reading View the code

About 90 minutes of reading.

Selected chapters

  1. 01Biological backgroundSequence, evolution, and protein structure.
  2. 03The EvoformerMSA and pair representations, attention, and updates.
  3. 10The Structure ModuleFrames, geometry, and Invariant Point Attention.
  4. 14A small training exampleFit a small model to a single protein structure.
View all sections

About the annotations

What’s in an annotation?

The explanations sit alongside code excerpts and figures, with links to the original paper and the full implementation.

  1. 01 / EXPLANATIONS

    The math and architecture

    Biological background, notation, and explanations of the main operations. Equations use the same symbols as the paper wherever possible.

  2. 02 / CODE

    PyTorch code

    Excerpts from a runnable implementation, with tensor shapes and references to the supplement. Each excerpt links to its source.

  3. 03 / FIGURES

    Interactive examples

    Controls for stepping through attention and inspecting coordinate frames, plus a 3D view of a protein during training.

About the annotations

BioTorch

Practice the operations
in PyTorch.

BioTorch has coding exercises for attention, protein geometry, and other operations used in biological AI. Write a function, run it, and check your results against the tests.

Practice on BioTorch