Chapter 5: Interpreting Structural Confidence: pLDDT, PAE & Disorder
Chapter 5: Interpreting Structural Confidence: pLDDT, PAE & Disorder
AlphaFold does not simply output a static PDB coordinate file; it provides rigorous, calibrated statistical confidence metrics for every atom and residue pair. Knowing how to interpret these confidence metrics is essential to avoid over-interpreting unconfident regions.
The Two Critical Confidence Metrics
- pLDDT (predicted Local Distance Difference Test): Measures local per-residue backbone confidence on a 0–100 scale (stored in the B-factor column of the output PDB).
- PAE (Predicted Aligned Error): Measures confidence in the relative distance and orientation between residue pairs (in Ångströms).
1. Understanding the Four pLDDT Quality Tiers
# The 4 pLDDT Quality Tiers (Stored in B-factor column):
# -----------------------------------------------------------------------------------------
# Score Range Color Band Structural Reliability / Interpretation
# -----------------------------------------------------------------------------------------
# > 90 Dark Blue Very High: Backbone and side chains highly accurate (crystal grade)
# 70 - 90 Cyan Confident: Correct global backbone fold; side chains may vary
# 50 - 70 Yellow Low: Low confidence; flexible loops or questionable topology
# < 50 Orange Very Low: Intrinsically Disordered Region (IDR) / unstructured ribbon
# -----------------------------------------------------------------------------------------2. Interpreting the Predicted Aligned Error (PAE) Matrix
The PAE matrix is a 2D heatmap where cell \((i, j)\) indicates the expected error in the position of residue \(x\) if the predicted and true structures are aligned on residue \(y\):
- Dark Green Off-Diagonal Blocks: Indicate well-defined rigid tertiary domains where relative positions are certain.
- High Error (Light/Red) Between Blocks: Indicates that two domains fold independently as rigid units, but the flexible linker connecting them allows them to move freely relative to each other. Do not interpret the relative inter-domain orientation as a fixed static state!
3. Visualizing pLDDT in PyMOL
# Color AlphaFold structure by pLDDT confidence bands
fetch AF-P04637-F1, type=pdb # Human p53
color blue, b > 90
color cyan, b > 70 and b <= 90
color yellow, b > 50 and b <= 70
color orange, b <= 50