How is a cluster representative sequence determined

[nooryoussef]

I used MMseqs2 to cluster sequences and am trying to understand how the representative sequence for each cluster is determined. Here’s a summary of the code I ran:

# Step 1: Convert the fasta file to a database (only needs to be done once for each fasta file)
mmseqs createdb fasta_file.fa queryDB

# Step 2: Cluster the database
mmseqs cluster queryDB queryDB_clus tmp --min-seq-id 0.75 -c 0.8 --cov-mode 1

# Step 3: Create a TSV file of the clusters
mmseqs createtsv queryDB queryDB queryDB_clus queryDB_clus.tsv

To investigate how the representative sequence is selected, I ran the following additional steps:

# Step 4: Compute distances between all sequences within a cluster
mmseqs alignall queryDB queryDB_clu queryDB_alnall

# Step 5: Create a TSV file of the alignment results
mmseqs createtsv queryDB queryDB queryDB_alnall queryDB_alnall.tsv

After analyzing the average identity and average score between sequences within each cluster, I found that neither of these metrics is highest for the sequence chosen by MMseqs as the representative. How is the representative sequence determined in this case?