Merge branch 'main' of https://huggingface.co/datasets/drake463/FireProtDB

README.md

@@ -8,7 +8,6 @@ configs:
     path: data/subsets/mutation_ddg/validation.parquet
   - split: test
     path: data/subsets/mutation_ddg/test.parquet
-
 - config_name: mutation_dtm
   data_files:
   - split: train
@@ -17,7 +16,6 @@ configs:
     path: data/subsets/mutation_dtm/validation.parquet
   - split: test
     path: data/subsets/mutation_dtm/test.parquet
-
 - config_name: mutation_binary
   data_files:
   - split: train
@@ -26,7 +24,6 @@ configs:
     path: data/subsets/mutation_binary/validation.parquet
   - split: test
     path: data/subsets/mutation_binary/test.parquet
-
 - config_name: mutation_lm
   data_files:
   - split: train
@@ -35,7 +32,6 @@ configs:
     path: data/subsets/mutation_lm/validation.parquet
   - split: test
     path: data/subsets/mutation_lm/test.parquet
-
 - config_name: protein_landscape_flat
   data_files:
   - split: train
@@ -44,4 +40,96 @@ configs:
     path: data/subsets/protein_landscape_flat/validation.parquet
   - split: test
     path: data/subsets/protein_landscape_flat/test.parquet
----
+license: cc-by-4.0
+language:
+- en
+tags:
+- chemistry
+- biology
+pretty_name: FireProtDB_2.0
+---
+
+# Dataset Card for FireProtDB_2.0
+
+Subsets of protein stability data for single-point mutants from FireProtDB, a comprehensive curated database.
+
+## Dataset Details
+
+Subsets of the FireProtDB database with train/validation/test splits:
+  1. 
+  2.
+  3.
+
+### Dataset Description
+
+This dataset contains hallucinated cyclic peptide scaffold structures (in CIF format) generated using AfCycDesign, a deep learning approach built on AlphaFold2 for de novo design of cyclic peptides. The scaffolds span peptide lengths of 7–16 residues and were generated through a hallucination procedure that simultaneously samples sequence and structure to produce well-ordered cyclic peptides. Structures in this dataset were filtered for high prediction confidence (pLDDT > 0.9) by AfCycDesign, representing 24,104 structurally diverse peptides predicted to fold into their designed conformations. These scaffolds serve as starting points for incorporating functional elements such as target-binding motifs for therapeutic applications.
+
+- **Curated by:** Zachary Drake / zacharydrake (at) g.ucla.edu
+
+### Dataset Sources
+
+- **Repository:** https://loschmidt.chaemi.muni.cz/fireprotdb/
+- **Paper:** Milos Musil, Simeon Borko, Joan Planas-Iglesias, David Lacko, Monika Rosinska, Petr Kabourek, Lígia O Martins, Mateusz Tataruch, Jiri Damborsky, Stanislav Mazurenko, David Bednar, FireProtDB 2.0: large-scale manually curated database of the protein stability data, Nucleic Acids Research, Volume 54, Issue D1, 6 January 2026, Pages D409–D418, https://doi.org/10.1093/nar/gkaf1211
+
+## Uses
+
+### Direct Use
+
+These hallucinated scaffolds are intended for use as starting structures in cyclic peptide binder design pipelines. Researchers can graft known binding motifs (e.g., hot loops from protein–protein interfaces) onto these scaffolds and redesign the remaining sequence using tools such as ProteinMPNN and Rosetta to create cyclic peptide binders against therapeutic protein targets. The scaffolds can also be used for benchmarking cyclic peptide structure prediction methods or as seed structures for further computational design campaigns.
+
+### Out-of-Scope Use
+
+These scaffolds are computational design models and have not all been experimentally validated. They should not be used as experimentally determined structures. The dataset is not intended for direct therapeutic use without extensive experimental characterization, including structural validation, binding assays, and stability testing. The scaffolds are composed of canonical L-amino acids only and do not natively support non-canonical amino acid design, though post-hoc substitution is possible.
+
+## Dataset Structure
+
+Each entry in the dataset includes a PDB file for hallucinated data representing the 3D atomic coordinates of a hallucinated cyclic peptide scaffold. Scaffolds range from 7 to 16 residues in length and are organized by peptide size (sets 7–16). Each scaffold represents a unique structural cluster identified through torsion bin-based clustering, where each residue is assigned a bin (A, B, X, or Y) based on its φ, ψ, and ω backbone torsion angles. All scaffolds in this dataset were predicted to fold into their designed structures with high confidence (pLDDT > 0.9) by AfCycDesign.
+Experimental entries include CIF files.
+
+#### Data Collection and Processing
+
+Scaffolds were generated using the AfCycDesign hallucination pipeline, which modifies AlphaFold2 with a custom cyclic offset matrix for relative positional encoding. For each peptide size (7–13 residues initially, later extended to 16), 48,000 hallucinated models were generated. Structures were clustered using a torsion bin-based approach, and the highest-confidence member from each cluster (pLDDT > 0.9) was selected. The hallucination is guided by losses that optimize prediction confidence metrics (pLDDT and predicted alignment error) and the number of intramolecular contacts. The pipeline was implemented within the ColabDesign v1.1.2 framework.
+
+## Citation
+
+**BibTeX:**
+
+```bibtex
+@article{10.1093/nar/gkaf1211,
+    author = {Musil, Milos and Borko, Simeon and Planas-Iglesias, Joan and Lacko, David and Rosinska, Monika and Kabourek, Petr and Martins, Lígia O and Tataruch, Mateusz and Damborsky, Jiri and Mazurenko, Stanislav and Bednar, David},
+    title = {FireProtDB 2.0: large-scale manually curated database of the protein stability data},
+    journal = {Nucleic Acids Research},
+    volume = {54},
+    number = {D1},
+    pages = {D409-D418},
+    year = {2025},
+    month = {11},
+    abstract = {Thermostable proteins are crucial in numerous biomedical and biotechnological applications. However, naturally occurring proteins have evolved to function in mild conditions, and laboratory experiments aiming at improving protein stability have proven laborious and expensive. Computational methods overcome this issue by providing a cheap and scalable alternative. Despite significant progress, their reliability is still hindered by the availability of high-quality data. FireProtDB 2.0 (http://loschmidt.chemi.muni.cz/fireprotdb) is a large-scale database aggregating stability data from multiple sources. The second version builds upon its predecessor, retaining its original functionality while introducing a new approach to data storage and maintenance. The new scheme enables the introduction of both absolute and relative data types connected with measurements of wild-types, mutants, protein domains, and de novo designed proteins. Furthermore, while the original database was limited to single-point mutations, more complex data such as insertions, deletions, and multiple-point mutations are now available. As a result, the inclusion of large-scale mutagenesis has increased the size of the database from 16 000 to almost 5 500 000 experiments. Moreover, the updated abstract scheme is fully expandable with any new measurements and annotations without the need for any restructuring. Finally, the tracking of history together with fixed identifiers is in accordance with the FAIR principles.},
+    issn = {1362-4962},
+    doi = {10.1093/nar/gkaf1211},
+    url = {https://doi.org/10.1093/nar/gkaf1211},
+    eprint = {https://academic.oup.com/nar/article-pdf/54/D1/D409/65405634/gkaf1211.pdf},
+}
+```
+
+**APA:**
+
+Musil, M., Borko, S., Planas-Iglesias, J., Lacko, D., Rosinska, M., Kabourek, P., Martins, L. O., Tataruch, M., Damborsky, J., Mazurenko, S., & Bednar, D. (2026). FireProtDB 2.0: large-scale manually curated database of the protein stability data. Nucleic acids research, 54(D1), D409–D418. https://doi.org/10.1093/nar/gkaf1211
+
+## Glossary
+
+- **AfCycDesign**: The deep learning framework described in this paper that adapts AlphaFold2 for cyclic peptide structure prediction and design by introducing cyclic relative positional encoding.
+- **pLDDT**: Predicted Local Distance Difference Test — a per-residue confidence metric from AlphaFold2, ranging from 0 to 1, where higher values indicate greater prediction confidence.
+- **PAE**: Predicted Alignment Error — a metric from AlphaFold2 that estimates the error in the relative position of pairs of residues.
+- **Hallucination**: A de novo design approach that simultaneously generates sequence and structure by optimizing AlphaFold2 confidence metrics starting from a random sequence.
+- **Pnear**: A Rosetta-derived metric (0 to 1) indicating folding propensity; a value of 1 means the designed structure is the single lowest-energy conformation.
+- **CIF**: Crystallographic Information File — a standard file format for representing 3D molecular structures.
+- **Torsion bin clustering**: A method for grouping peptide structures by assigning bins (A, B, X, Y) based on backbone φ, ψ, and ω dihedral angles.
+
+## More Information
+
+The AfCycDesign code is implemented within the ColabDesign framework. For additional details on methods and supplementary data, see the full paper and its supplementary materials at https://doi.org/10.1038/s41467-025-59940-7.
+
+## Dataset Card Authors
+
+Jie Chen / jiechen7 (at) uw.edu, Zachary Drake / zacharydrake (at) g.ucla.edu, Adriana Hernandez Gonzalez / ahgonzalez (at) ucdavis.edu, Akshaya Narayanasamy / akshayanarayanasamy (at) gmail.com, Lina Maria Pena Pardo / linamp (at) umich.edu