MODEL LEARNING PATH
DiffDock
Original DiffDock predicts how a ligand should translate, rotate and change torsions inside a fixed receptor. A separate all-atom confidence network ranks the generated poses.
Loading the interactive page…
DiffDock made ligand docking a generative inference problem over translation, rotation and torsions, followed by learned pose ranking. Its framework directly motivated DiffDock-PP and DiffDock-L. The card year is ICLR 2023 (first preprint: 2022); original validation is computational pose prediction, not prospective binding experiments.
Chapters
Construct molecular context
Prepared ligand/receptor features and typed geometric relations feed the score network.
Build typed geometric relations for a ligand–receptor score network.
Predict motion scores
Equivariant readouts produce separate translation, rotation and bond-torsion scores.
Keep translation, rotation and torsion score outputs distinct.
Update the ligand pose
A manifold sampling step supplies increments; rigid motion, directed torsions and alignment apply them to coordinates.
Apply pose increments to the intended atoms and align the result.
Rank completed poses
A separately trained all-atom network and pooled confidence head assess candidate poses.
Aggregate candidate-pose features for a separate confidence prediction.
All exercises in this guide
- Masked Sum
- Masked Mean
- Stable vector norms, normalization, dot and cross features
- Masked pairwise distances and hard distance-bin targets
- Masked batched kNN graphs and neighbor gather
- Ragged segment sum and mean
- Typed radius, bond-prefix and sequence-neighbor graph assembly
- Real spherical harmonics from Cartesian polynomial tables
- Affine Projection
- Real-irrep Clebsch–Gordan paths and channel tensor products
- Origin-safe Bessel channels with a compact polynomial cutoff
- Radial-weighted irrep convolution with receiver aggregation
- Layer Normalization
- BatchNorm training, evaluation and functional running statistics
- Irrep-safe batch and instance normalization
- Ligand translation, rotation, bond torsion and confidence heads
- Gaussian forward corruption and score targets
- Wrapped torsion corruption, density and score normalization
- Batched rigid transforms
- Stable SO(3) exponential and logarithm maps
- SO(3) heat-kernel scores and supplied-variate angle sampling
- Masked sums, means and weighted reductions
- Shared rigid coordinate augmentation and Haar rotations
- Three-space docking initialization and reverse diffusion step
- Sequential directed-bond torsion rotations
- Weighted proper rigid alignment
- ReLU
- Feedforward Block
- Activations and gated feed-forward transitions
- Affine projection and embedding lookup
- DiffDock three-space score-matching loss