User Manual

1. Overview

The LGNM Server (Learned Graph Network Model) predicts per-residue protein flexibility from a single static structure. It combines:

• Elastic Network Models (ENM) — coarse-grained normal mode analysis
• Graph Neural Networks (GNN) — learned spring constants and flexibility scores
• Quantum-inspired Neural PDE — efficient Hamiltonian dynamics with O(K) operator cost

2. Input Options

Fetch from RCSB

Enter a 4-character PDB ID (e.g., 6LU7, 1UBQ). The server downloads the structure from the RCSB Protein Data Bank.

Upload PDB File

Upload a .pdb or .cif file (max 50 MB). Drag-and-drop is supported.

Chain Selection

After loading, all available chains are displayed. Select one chain for analysis. Multi-chain support is available — run each chain separately to compare dynamics.

3. Understanding Output

Per-Residue Metrics

Plots

• Flexibility Profile — RMSF vs residue index with top-20% flexible regions shaded
• Flexibility Map — bar chart coloured by flexibility with sequence annotation
• Stiffness Heatmap — per-residue stiffness (1D strip)
• Conformer Spread — ENM-sampled conformational variability
• Normal Modes — top 5 slowest modes showing collective motions

Downloads

• Colored PDB — B-factors replaced with flexibility scores (0–100)
• CSV — tabular flexibility data for all residues
• JSON — complete results including modes and conformer data

4. 3D Viewer Guide

The interactive viewer uses Mol*.

Mouse Controls

Toolbar

• Representations: Cartoon, Surface, Backbone, Ball & Stick, Spacefill
• Reset: Return to default camera position
• Auto Rotate: Continuous slow rotation
• Background: Toggle white/black background
• Screenshot: Export high-resolution PNG

Sidebar Controls

Use the collapsible sidebar to change coloring modes, color palettes, highlight flexible/rigid residues, and download results.

5. Interpreting Flexibility

What does high flexibility mean?

• Residues with flexibility > 0.7 are typically in loops, termini, or disordered regions
• These regions often participate in binding, allostery, and conformational changes
• High flexibility does not mean the structure is incorrect — it reflects genuine dynamics

What does low flexibility mean?

• Residues with flexibility < 0.3 are typically in the protein core
• α-helices and β-sheets tend to be more rigid than loops
• Buried hydrophobic residues (high burial score) are usually rigid

Amino Acid Trends

Comparing with Experimental Data

Predicted RMSF correlates with:

• Crystallographic B-factors (Pearson r ≈ 0.6–0.8 for well-resolved structures)
• NMR order parameters S² (inverse correlation)
• MD simulation RMSF (Pearson r ≈ 0.7–0.9)

6. Methods & Theory

Gaussian Network Model (GNM)

The GNM treats the protein as a network of Cα atoms connected by harmonic springs within a cutoff distance d_c. The Kirchhoff matrix Γ encodes the connectivity, and its pseudoinverse gives the covariance of fluctuations.

Modified ENM (mENM)

Instead of uniform spring constants, mENM uses distance-dependent coupling: γ_ij ∝ 1/r^p_ij, with p = 2 by default. This gives more realistic flexibility at protein surfaces.

Learned GNM (LGNM)

A graph neural network predicts per-edge spring constants and per-residue flexibility scores from sequence and structural features. Features include amino acid identity (one-hot), burial depth, secondary structure, and pLDDT (when available from AlphaFold models).

Quantum-Inspired PDE

The Hamiltonian dynamics module uses K operator types acting on M edges, achieving O(K) cost versus classical matrix diagonalisation at O(N³). This enables efficient analysis of very large proteins.

7. FAQ