Supported Formats for Predicted Alignment Error Matrices

Predicted Alignment Error matrices are square matrices produced as part of deep-learning based 3D-structure prediction pipelines such as AlphaFold. They can be imported via Jalview's structure chooser GUI and the Command Line Interface. See Working with PAE Matrices for information on how they are visualised and analysed in Jalview.

Supported Formats

Jalview supports import of PAE matrix data as provided by the EBI-AlphaFold database. This resource provides PAE matrices as a JSON files structured in one of the following ways:

	# Version 1 format PAE file - deprecated 28th July 2022
	{
		residue1:[1,1,... total number of residues]
		residue2:[1,2,... total number of residues]
		distance:[0.1,0.3,... list of PAE matrix elements as doubles]
	}
  

  	# Version 2 format PAE file - see https://alphafold.ebi.ac.uk/faq
  	{
  		max_predicted_alignment_error: 4.0, # may also be max_pae
  		predicted_alignment_error: [[1,2,0,0,3,...],...] # may also be pae
  	}
  

Variants of the version 2 format include using 'pae' instead of 'predicted_alignment_error' in the names of keys. Jalview copes both.
Once imported, Jalview stores PAE matrices as float arrays along with any associated tree and partition set resultant from clustering the matrix.

PAE Matrix import support was added in Jalview 2.11.3