Update README.md

README.md

@@ -237,9 +237,17 @@ with (snapshot / "metadata.csv").open(newline="") as handle:
 print(out_dir)
 ```
 
-## Notes
+## Citation
 
-- This is a raw mirror, not a cleaned or reformatted training set.
-- The alignment tree is preserved under its original relative paths inside the tar shards.
-- Tar shards are uncompressed. The goal is predictable extraction and straightforward random access to individual members.
-- Check the upstream OpenProteinSet/OpenFold data terms and cite the original resources as appropriate for your use case.
+```
+@inproceedings{ahdritz2023openproteinset,
+  title     = {OpenProteinSet: Training Data for Structural Biology at Scale},
+  author    = {Ahdritz, Gustaf and Bouatta, Nazim and Kadyan, Sachin and Jarosch, Lukas and Berenberg, Daniel and Fisk, Ian and Watkins, Andrew M. and Ra, Stephen and Bonneau, Richard and AlQuraishi, Mohammed},
+  booktitle = {Advances in Neural Information Processing Systems 36: Datasets and Benchmarks Track},
+  year      = {2023},
+  url       = {https://openreview.net/forum?id=gO0kS0eE0F},
+  eprint    = {2308.05326},
+  archivePrefix = {arXiv},
+  primaryClass  = {q-bio.BM}
+}
+```