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.