Digging into the 3D Structure Predictions of AlphaFold2 with Low Confidence: Disorder and Beyond

Area of Science:

• Structural biology
• Computational biology
• Bioinformatics

Background:

• AlphaFold2 (AF2) provides highly accurate 3D protein structure models.
• The AF2 pLDDT score correlates with residue-wise protein disorder.
• pyHCA tool identifies foldable protein segments and their order/disorder ratio.

Purpose of the Study:

• To combine AF2 predictions with pyHCA for analyzing protein structure and disorder.
• To investigate long foldable segments (>30 amino acids) with soluble domain characteristics.
• To compare AF2 segments with low and high pLDDT values.

Main Methods:

• Utilized AF2 predictions from the AlphaFold Protein Structure Database (AFDB v1) for 21 reference proteomes.
• Applied pyHCA to identify and characterize foldable segments within AF2 predictions.
• Analyzed segments based on pLDDT scores to differentiate between ordered and disordered regions.

Main Results:

• Identified long foldable segments exhibiting characteristics of soluble domains.
• Characterized segments with consistently low pLDDT values, suggesting disorder.
• Compared these with segments of high pLDDT values, revealing conditional order.
• Highlighted cases missed by AF2 due to shallow multiple sequence alignments or undocumented folds.

Conclusions:

• Combining AF2 and pyHCA effectively reveals cryptic structural features in whole proteomes.
• This integrated approach refines predictions for various types of protein disorder.
• The study enhances understanding of protein structural flexibility and prediction limitations.