Novel in-house RAPTOR phase

RIGOR: Post-Docking Triage for Peptide Binder Discovery

RIGOR-Lite is RAPTOR's in-house deterministic scoring phase. It evaluates docked peptide candidates after DiffDock and before GROMACS molecular dynamics, filtering weak or physically implausible binders before expensive simulation. Unlike generic docking scores, RIGOR evaluates whether a candidate forms a resilient interface, occupies meaningful receptor patches, avoids excessive conformational penalty, and remains sequence-developable.

A proprietary RAPTOR scoring phase that converts docked peptide poses into an explainable advance / review / reject decision before molecular dynamics.

Example output

Candidate

binder_001

ADVANCE

Final Score

78.4/ 100

Confidence

68.5%0–100 scale

CIRS-lite81.2
HARP-lite76.5
ECHO-lite72.0
BALM-lite84.0

Illustrative values for a single docked candidate.

Pipeline placement

Where RIGOR sits in the pipeline

PDBFixer
RFdiffusion
DiffDock
RIGOR-LiteIn-house
Ligand Prep
GROMACS MD
ADMET-AI
eTox
Final Ranking
Inserted after docking, before MD

RIGOR is placed after DiffDock because it requires a receptor-peptide docked pose. It is placed before GROMACS because molecular dynamics is computationally expensive. RIGOR therefore acts as a low-cost gate that prevents poor docked poses from consuming downstream simulation resources.

Motivation

Why RIGOR is needed

Docking scores are not enough

Docking can produce poses that look numerically favorable but lack stable or biologically meaningful receptor-peptide contacts.

MD is expensive

Running molecular dynamics on every generated binder wastes computational resources when many poses are structurally weak.

Peptide binders are flexible

Peptides can appear favorable in a single pose but may pay a large conformational cost or fail to preserve contacts under small perturbations.

Downstream MD cost

Illustrative
Without RIGOR100% MD cost
With RIGOR~35% MD cost
Advance → MDReviewReject (skipped)All poses (no gate)

RIGOR fills this gap by adding an interpretable post-docking intelligence layer between docking and molecular dynamics.

Hybrid inputs

RIGOR input model

Input 1

Receptor PDB

Source:
PDBFixer — exact file used during DiffDock/SMINA docking
Purpose:
Receptor coordinates in the same frame as the docked pose
Input 2

Docked peptide pose

Source:
Selected ranked DiffDock SDF (e.g. rank1_confidence-3.62.sdf)
Purpose:
Docked spatial coordinates — preserved exactly, never re-docked
Input 3

Original peptide sequence

Source:
RFdiffusion metadata
Purpose:
True peptide identity (not inferred from the SDF)
RFdiffusion sequence+DiffDock docked .sdf+Exact docking receptor .pdbRIGOR-Lite scoring

Exact-coordinate contract

RIGOR must receive the exact receptor file used during DiffDock — not a re-extracted chain — and the exact selected ranked DiffDock SDF (or a coordinate-preserving PDB converted from that same SDF). The peptide sequence always comes from RFdiffusion metadata; RIGOR never infers it from an SDF-derived ligand residue such as LIG, UNL, MOL, or X.

Inside RIGOR-Lite

Component modules

Module

CIRS-lite

Contact Integrity Resilience Scorer

Measures whether receptor-peptide contacts remain stable under small pose perturbations.

  • contact persistence
  • Jaccard contact retention
  • clash penalty
  • resilience score
Module

HARP-lite

Hotspot-Aligned Receptor Patch Score

Evaluates whether the peptide contacts coherent and chemically meaningful receptor surface patches.

  • contact density
  • chemistry match
  • patch coherence
  • anchor residue fraction
Module

ECHO-lite

Entropic Commitment and Helical Order Proxy

Estimates whether the peptide is too flexible or sufficiently preorganized for binding.

  • contact fraction
  • flexibility penalty
  • internal constraint score
  • length/preorganization score
Module

BALM-lite

Bound-state Aware Liability Matrix

Screens peptide sequence liabilities related to developability, synthesis burden, and stability.

  • deamidation motifs
  • oxidation-prone residues
  • hydrophobic clusters
  • cysteine complexity
  • sequence liability score
Module

FUSE-lite

Fusion and Uncertainty Scoring Engine

Combines all component scores into a single final RIGOR score, confidence value, and decision.

  • final score
  • confidence
  • ADVANCE / REVIEW / REJECT decision

How RIGOR fuses signals

Score formula

RIGOR-Lite Score =
  0.40 × CIRS-lite
+ 0.25 × HARP-lite
+ 0.20 × ECHO-lite
+ 0.15 × BALM-lite
CIRS-lite · 40%
HARP-lite · 25%
ECHO-lite · 20%
BALM-lite · 15%

The current weights are heuristic starting values and can later be recalibrated using benchmark data or internal experimental results.

From score to action

Decision logic

85–100Very strong candidate
75–84Advance
55–74Review
35–54Weak
0–34Reject

No-contact rule

If no receptor-peptide interface contacts are detected, the candidate is automatically rejected or flagged as input-error depending on the diagnostic result.

Confidence scale

Confidence is reported on a 0–100 scale. A value of 40 means 40% confidence — the number is already a percent and never multiplied again.

Mock RIGOR report

Visual diagnostic dashboard

Final score

78.4/ 100

Decision: ADVANCE

Component scores

CIRS-lite81.2
HARP-lite76.5
ECHO-lite72.0
BALM-lite84.0

Weighted contributions

CIRS × 0.4032.5
HARP × 0.2519.1
ECHO × 0.2014.4
BALM × 0.1512.6

Interface diagnostics

Atom contacts
184
Residue contacts
31
Receptor contact residues
18
Peptide contact residues
13
Clashes
2
Min Rx–Pep distance
2.41 Å

Warnings

  • Invalid sequence (X)
  • SDF-derived ligand representation
  • Coordinate-frame mismatch
  • No peptide atoms parsed

Novelty

What makes RIGOR novel inside RAPTOR?

RIGOR is the only fully in-house phase in the RAPTOR pipeline.
It is not another external docking engine or generative model.
It integrates multiple interpretable structural and sequence criteria into one scoring gate.
It is designed specifically for peptide binders.
It is CPU-only and does not require GPU inference.
It produces auditable reports instead of black-box predictions.
It reduces wasted downstream MD by filtering weak poses immediately after docking.
It connects RFdiffusion identity with DiffDock geometry in a single explainable triage layer.
“RIGOR transforms RAPTOR from a pipeline that only integrates external tools into a pipeline with a proprietary decision-making phase.”

RIGOR output

What you get back

≥ 75

ADVANCE

Pose is sent forward into ligand prep and GROMACS MD.

55–74

REVIEW

Pose is borderline — you decide whether to continue or stop.

< 55

REJECT

Pose is filtered out before expensive simulation.

Score breakdown

100%weight
  • CIRS-lite40%
  • HARP-lite25%
  • ECHO-lite20%
  • BALM-lite15%

Handoff diagram

DiffDock pose

.sdf (exact)

Receptor

.pdb (docking frame)

RFdiffusion

sequence

RIGOR-Lite

Gate decision

advance / review / reject

Component scores

CIRS · HARP · ECHO · BALM

RIGOR receives the exact docked SDF, the exact docking receptor, and the RFdiffusion sequence — and returns a gate decision plus four explainable component scores.

Honest scope

Scientific limitations

RIGOR-Lite is an early-stage triage method, not a direct experimental binding-affinity predictor. It does not replace molecular dynamics or wet-lab validation. It does not fully model induced fit, explicit water networks, metals, glycans, noncanonical residues, or true binding free energy. Its purpose is to provide fast, interpretable prioritization before expensive downstream simulation.

RIGOR turns docking output into a decision

By combining contact resilience, receptor patch relevance, entropy awareness, and sequence developability, RIGOR provides RAPTOR with a novel, explainable, and computationally efficient post-docking triage phase.