REACH-Ab | Dynamic Interface Edge Field

Key Scientific Question

Can an antibody generator form the interface relation before the paratope is fully stabilized?

Antibody CDR generation is not only a problem of producing plausible loop geometry. Under epitope conditioning, the model must learn which paratope residues should form compatible relations with the antigen surface, and use those relations to shape sequence and structure together.

Core Innovation

Dynamic Interface Edge Field

REACH-Ab treats cross-interface edges as learnable, evolving relation variables instead of passive kNN or distance-threshold links. Candidate epitope-paratope relations compete, persist, or disappear across refinement rounds, then active edges carry relation states into interface message passing.

Half-edge handshake Two interface sides propose compatible relation evidence.

Edge memory Candidate relations keep history across generation steps.

Lifecycle scoring Relations can be born, maintained, suppressed, or marked uncertain.

Preparing structure view...

Mechanism Figures

How REACH-Ab turns epitope guidance into evolving interface relations

These figures summarize the scientific story behind the viewer: REACH-Ab first defines the antibody design task at the complex level, then maintains candidate epitope-paratope relations as dynamic states, and finally feeds selected relation states back into local message passing.

Framework of REACH-Ab with input initialization, global graph, local graph, iterative edge-node co-evolution, and outputs — Framework Overview

From masked CDR residues to generated antibody sequence and complex structure

The framework initializes masked CDR residues from an antibody-antigen complex, builds global and local graphs, and iteratively couples SE(3) message passing with a local Dynamic Interface Edge Field. The key idea is that candidate interface edges are explored, updated, and selected before they guide local coordinate refinement and sequence generation.

Schematic of the Dynamic Interface Edge Field moving from candidate edge exploration to stable interface edge formation — Core Concept

A dense relation field becomes a sparse active interface graph

REACH-Ab maintains many possible epitope-paratope contacts instead of committing only to current geometric neighbors. During refinement, half-edge evidence and lifecycle control let some relations stabilize, some compete, and some die out. This is the mechanism that reduces epitope guidance lag: antigen-facing contact hypotheses remain available before the paratope geometry is fully reliable.

Detailed implementation flow of candidate edge construction, half-edge probing, dynamic memory, lifecycle heads, active edge selection, and message passing — Implementation Flow

Candidate edges carry memory, compatibility, uncertainty, and lifecycle signals

At each refinement step, residue states and anchors construct a directed candidate edge pool. Pair-level features are updated through source-side and destination-side half-edge probing, then fused into persistent edge memory. Distance, occupancy, birth, death, persistence, and uncertainty heads produce an active edge score; selected edge states then become attributes in local message passing and are carried to the next refinement step.

Interactive Mechanism Playground

Feel the epitope guidance lag before reading about it

This toy generation sandbox turns the central mechanism into something visitors can manipulate. The left side rebuilds interface edges only from the current node geometry, while the right side maintains a broader edge field with memory and compatibility evidence before selecting active message-passing edges.

Early coordinates control geometry-kNN edges immediately; REACH-Ab keeps candidate contact hypotheses available while compatibility and memory evidence accumulates.

Process-Level Readout

Edges should guide nodes before nodes are reliable.

Early CDR geometry is noisy, so nearest-neighbor edges can expose the wrong antigen contacts. The edge field keeps native-like candidates visible until they can become active edges.

Native contact visibility -

Active native edges -

False-positive reinforcement -

Guidance timing -

CDR geometry uncertainty Edge memory strength Show faint candidate field

Edge-Field-Guided Generation

Watching contact hypotheses form during generation

The main-paper mechanism claim is temporal: REACH-Ab should expose useful epitope-paratope contact hypotheses before the final CDR geometry is settled. These 57 trajectories compare a geometry-kNN proxy with the REACH-Ab edge field for the same benchmark cases shown in the PDB viewer.

Animated comparison of geometry-kNN proxy edges and REACH-Ab edge-field edges for 4FQJ — 4FQJ edge-field trajectory

The left panel reconstructs edges after provisional node geometry appears. The right panel shows REACH-Ab maintaining and activating candidate edge states during refinement, so contact evidence can participate while the paratope is still being formed.

Mechanism Reading

What the trajectory is meant to prove

If epitope guidance lag is real, the important evidence is not only the final contact map. The process should show whether native-like contacts become visible, active, and stable early enough to guide the next node update.

Contact visibility Whether native epitope-paratope pairs remain accessible in the candidate field.

Contact activation Whether useful candidate edges become message-passing edges during refinement.

Contact emergence Whether stable contacts arise before the final paratope geometry is fully determined.

False-positive control Whether unsupported geometry-induced edges are prevented from reinforcing wrong contacts.

The gray dashed edges on the left are reactive: geometry first, edge afterward. The purple edges on the right are generative variables: candidate relations accumulate compatibility, memory, lifecycle, and uncertainty evidence, then feed selected edge states back into CDR sequence and structure updates.

Paper to Code

Method equations linked to their implementation

This section merges the main-paper REACH-Ab method with the supplementary Model Architecture and Implementation Details and Theoretical Analysis. Each formula below is clickable: selecting it opens the concrete PyTorch implementation used by the current codebase.

Architecture Thread

REACH-Ab starts from a broad directed epitope-paratope candidate pool, encodes pair geometry and endpoint states, injects two-sided half-edge evidence, stores candidate-level memory, predicts lifecycle and uncertainty signals, then selects a sparse active interface graph for local message passing.

Theoretical Thread

The supplementary analysis explains why this matters: a broad candidate space preserves early contact accessibility, vector edge states are more expressive than scalar scores, half-edge probing generalizes one-sided scoring, and memory allows delayed activation from repeated weak evidence.

Implementation

How REACH-Ab turns epitope guidance into evolving interface relations

From masked CDR residues to generated antibody sequence and complex structure

A dense relation field becomes a sparse active interface graph

Candidate edges carry memory, compatibility, uncertainty, and lifecycle signals

Feel the epitope guidance lag before reading about it

Step 0: uncertain paratope hypothesis

Watching contact hypotheses form during generation

4FQJ edge-field trajectory

Method equations linked to their implementation

Architecture Thread

Theoretical Thread