CONCLUSION
The factors that would have significantly contributed to the results of
the CASP15 were explored. The structure-building step was performed
using AlphaFold2 and AlphaFold-multimer, indicating that the AF2 series
is a state-of-the-art structure prediction tool. A detailed protocol
analysis suggests that extensive MSA construction and manual
interventions considering interfaces were the main factors for the high
performance of the CASP15. However, there was at least one target in
which other teams received more sequences for MSA; thus, there is still
room for improvement in database construction or sequence similarity
search steps. The performance could have been higher in the multimer
categories, and I could not predict a sufficient model in most cases
when no evolutionarily related sequences were observed. Therefore,
developing a more effective protocol to address these issues is
required. In addition, transporting the manual intervention process to
automation will be an upcoming issue because manual intervention should
be minimal.