extHomFam 2: large-scale benchmark for protein multiple sequence alignments

extHomFam 2 was constructed by combining Homstrad reference alignments (March 2020) with Pfam 33.1 complete families (NCBI variant). Homstrad entries with less than 3 reference sequences and those pointing to dead Pfam families were discarded. The resulting benchmark was divided into subsets depending on the family size N:

subset
N range
# families

small
[200, 10 000)
86

medium
[10 000, 40 000)
95

large
[40 000, 100 000)
83

xlarge
[100 000, 250 000)
67

huge
[250 000, 3 000 000)
62

The directories in the archive correspond to the names of the subsets, while the reference alignments are located in 'ref' folder.