Benchmarking

Benchmarking on a few datasets:

1. Dataset from the previous benchmark paper https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6729087/
Includes large ligands and tests the incremental approach success.
2. Ensemble datasets.
Would be nice to show performance on 2 more ensemble datasets (estrogen receptor goes separately)
- CDK2 dataset
- 453 crystal structures in the PDB
- representative ensemble of 5 structures (using EnGens)
- 383 unique ligands
- EnGens analysis: https://colab.research.google.com/drive/1tq_vcp6XBakBPUJ_EavyasVyaWVloTn5
- PI3K dataset
- ensemble from the EnGens paper (recreate)
- look into ligands
- ER dataset
- ~400 crystal structures
- ~300 unique ligands
- ensemble 3 structures

Start by reviewing the previous benchmark paper and the linked EnGens analysis for the CDK2 dataset, then identify how the CDK2, PI3K, and ER ensembles and ligands should be assembled. Done means benchmark results are produced for the prior-paper dataset and the requested ensemble datasets, with the ER dataset handled separately.

Written by the indexing model from the issue text.