AlphaFold2 Pre-Processing

AlphaFold2 Pre-Processing

Let's talk a bit about how AlphaFold2 go from a protein FASTA file to a full structure prediction.

Searching for Similar Sequences

• This query sequence is compared to:
  • UniRef90 database to find similar sequences
  • PDB70 to find similar structures
• Sequences that are too similar to our query are filtered out so that we don’t just build a replicate based on that sequence
• These sequences are arranged as an MSA

Multiple Sequence Alignment (MSA)

• An MSA is an array of sequences
• These sequences are aligned with one another as to best match similar regions
• These sequences don’t always line up perfectly and as such we see:
  • Conserved positions: where the letter does not change
  • Coevolved positions: where the letter will change with another letter
  • Specificity Determining positions: where the letter is consistently different

Why is an MSA Useful In Structure Prediction?

• The theory is that residues that coevolve are generally close to each other in the protein’s folded state
• So, by assessing what residues change together we get an idea of where they might be spatially!

MSA Embedding

• An MSA is still essentially an array of letters
• To be more computer friendly these letters are embedded as numbers using their positional information
• AlphaFold embeds these letter values as numeric ones and terms this the MSA representation

Embedding Example

• Take for example the sentence “I ate an apple and played the piano”
• This string is embedded by positional information.
• e.g. ate was the second word so there is a 1 in the second column at row “ate”

Pair Representation

• Similar Structures were also queried for using our protein sequence.
• These structure files (A.K.A Crystallographic Information Files (CIF)) contain 3D coordinates for a protein’s atoms in space
• These coordinates are used to initialize a pairwise distance matrix between residues that AlphaFold calls the pair representation