Detecting Mirror Images in Coarse-Grained Simulations

Created on April 23, 2026

2026   ·   coarse-grained-simulation   protein-folding   chirality   analysis   ·   research-notes

In coarse-grained structure-based folding simulations, a protein can sometimes fold into a global mirror-image structure rather than the physically correct native fold. These mirror-image states are typically artifacts for an all-L polypeptide and should be removed from downstream analysis.

Purpose

This workflow is designed to:

  • Flag trajectories trapped in a global mirror-image basin
  • Distinguish them from native-like and non-mirror misfolded trajectories
  • Remove only global mirror-image artifacts

Why mirror-image states appear in CG models

In many structure-based CG models:

  • Native contact topology is preserved
  • Global fold geometry is approximated
  • Chirality is not strictly enforced

As a result, a mirrored fold can still be compact, stable, and even have high Q, despite being physically invalid for all-L proteins.

Detection principle

A global mirror-image trajectory is expected to satisfy all of the following:

  • Folded or near-folded (Q is high)
  • Low agreement with native chirality (inverted handedness)
  • Geometrically closer to reflected native than to native

So detection combines:

  • Q (folded-state filter)
  • Global chirality agreement score vs native
  • RMSD_native
  • RMSD_reflected

Why Q alone is not enough

Q measures native-contact similarity but does not encode handedness. Therefore, mirror-image conformations can still show high Q.

Use the same late-time window for all time-averaged quantities below (for example, last 10 ns).

Step 1: Keep folded trajectories

For each trajectory, compute late-window mean ⟨Q⟩_last and keep:

⟨Q⟩_last > 0.6

Step 2: Compute global chirality agreement

Compute a scalar chirality agreement score against the native chirality structure:

  • Larger values: better native handedness agreement
  • Smaller values: mirror-like inversion

Threshold used here:

chirality < 0.2

Step 3: Build reflected native reference

Reflect the native structure by flipping one coordinate axis (for example, x -> -x). One axis flip is sufficient because reflections differ only by rotation.

Step 4: Compare RMSD to both references

Compute late-window means of:

  • RMSD_native (rotation-only alignment to native)
  • RMSD_reflected (rotation-only alignment to reflected native)

Important: use proper rotations only (det = +1). Reflection should enter through the reflected reference structure, not through allowing improper alignment transforms.

Final classification rule

Classify trajectory as mirror-image artifact if and only if:

(⟨Q⟩_last > 0.6) and (chirality < 0.2) and (RMSD_reflected < RMSD_native)

This three-way AND is important to reduce false positives from any single metric.

Optional refinement

Define:

mirror_score = RMSD_reflected / RMSD_native

and add a stricter cutoff such as:

mirror_score < 0.8

to remove borderline cases.

What this intentionally keeps

  • Unfolded trajectories (⟨Q⟩_last <= 0.6)
  • Local chirality defects (not global inversion)
  • Near-native misfolded states closer to native than reflected native

These classes can still contain meaningful physics and should remain available for analysis.

Summary

Mirror-image folds can look folded in CG simulations, so filtering should not rely on Q alone. A robust practical filter is the joint rule:

(⟨Q⟩_last > 0.6) and (chirality < 0.2) and (RMSD_reflected < RMSD_native)

This removes global mirror artifacts while preserving informative non-mirror misfolded states.