Fig. 9
From: Protein language models can capture protein quaternary state
Comparison of success of qs predictions of different methods: Structure-based PISA, EPPIC, and sequence-based QUEEN. A. Success (y-axis) for different qs (x-axis) for the different approaches. PISA and EPPIC predominantly outperform QUEEN. Note that PISA and EPPIC use not only the structure but also the full crystal information, and they only choose from the available options within the crystal lattice. B and C. Success (y-axis) colored according to performance by PISA and QUEEN for different qs, highlighting the significant overlap, but also the additional potential contribution of QUEEN. The division between B and C is for clarity of the y-axis scale, B showing the larger classes 1–4, 6, and C showing the smaller classes 5, 7–24