A real example: what you get
15 slides for a 20-minute talk · “Highly accurate protein structure prediction with AlphaFold.”
- Each point with its passage of the paper
- Earlier work and what came next, strengths and limits stated by the authors
- Where the paper stands: published, corrected, newer version
- 5 figures of the paper, with their credit
- 5 hard questions to prepare
- Presenter view, PowerPoint, Beamer and a link to share
A slide
Can evolutionary information and geometric inductive biases support near-experimental structure prediction? · Earlier orientation-based prediction combined learned restraints with Rosetta optimization. · AlphaFold instead predicts heavy-atom coordinates directly from sequence and homologous alignments.
Context and central question
Its source
The AlphaFold network directly predicts the 3D coordinates of all heavy atoms for a given protein using the primary amino acid sequence and aligned sequences of homologues as inputs (Fig. 1e; see Methods for details of inputs including databases, MSA construction and use of templates).
— Highly accurate protein structure prediction with AlphaFold.
The first results
-
Highly accurate protein structure prediction with AlphaFold
-
Context and central question
Can evolutionary information and geometric inductive biases support near-experimental structure prediction?
-
Joint evolutionary and geometric reasoning
Evoformer maintains interacting MSA and residue-pair representations.
-
Coordinate generation and refinement
Residue frames prioritize backbone orientation while temporarily allowing broken peptide geometry.
-
Training and self-distillation
Supervised structures had maximum release date 30 April 2018.
-
Evaluation design and controls
CASP evaluates predictions against recently solved, undisclosed structures.
-
CASP14 accuracy
Median backbone r.m.s.d.95 was 0.96 Å, versus 2.8 Å for the next-best method.
-
Generalization and confidence
High accuracy transferred to recent PDB submissions evaluated as full chains.
-
Where performance remains conditional
Accuracy decreases substantially below approximately 30 aligned sequences.
-
Take-home message and broader comparison
AlphaFold combines learned structural inference with physical and geometric inductive biases.
5 more results hidden
See a complete example, explained step by step →
The paper and its summary →
