3 | CONCLUSION
This paper summarizes the technical aspects of new prediction categories
in CASP15 and can serve as a reference point for future predictors. The
systems for handling RNA/ligand/multimer-EMA predictions in CASP15 were
implemented shortly before the start of the experiment and, as such,
were not extensively tested. Based on the feedback from CASP15 assessors
and predictors, we plan to improve the system by further automating the
process of verifying RNA and ligand submissions and ensuring better
compliance of models with the required format. This investment of time
and effort should pay off at the assessment stage with a more seamless
evaluation of predictions. Since the PDB format is becoming obsolete, we
will motivate predictors and developers of the evaluation measures used
in CASP to support the PDBx/mmCIF format in their tools and methods. For
the RNA category, we will work to further incorporate and standardize
evaluation measures which assess the base-base interaction network. This
will require setting community-accepted standards in identification and
annotation of RNA interactions22. Further, we will
work to improve measures comparing predictions to raw data to assess
low-resolution targets. For the ligand prediction category, we need to
improve the set of targets with more interesting ligand modeling
challenges, possibly by tapping into unreleased structures from
pharmaceutical companies. We will also improve the acceptance system by
validating the identity of submitted ligands in a more robust manner,
and we would like to consider introducing metrics for self-assessment of
the accuracy of ligand poses within the receptor. For evaluating
ensembles of alternative conformations, we will explore acceptance of
multi-model entities, with optional probability weights, as submissions.