Double dynamic programming
Taylor & Orengo (1989), Taylor (1997)
Need not fix one alignment, but makes high-level scoring matrix based on quality of fit for a set of superpositionings (low-level matrices)
- each low-level alignment assumes one pair (e.g., Ai paired with Bj)
- superpose the structures based on the two
- align the complete structures based on this superpositioning
- scores obtained propagated to high-level DP scoring matrix