Binding Affinity · LiteFold
SKEMPI2
Experimental benchmark of protein-protein binding affinity, kinetics, and thermodynamics changes upon mutation.
What SKEMPI2 is
SKEMPI v2 is a manually curated benchmark detailing experimentally measured changes in protein-protein binding affinity, kinetics, and thermodynamics resulting from mutation. It focuses on structurally resolved protein-protein interactions, linking single and multiple mutations directly to Protein Data Bank (PDB) complexes alongside literature-derived binding parameters.
The dataset addresses the challenge of validating and developing predictive algorithms for macromolecular interfaces. Computational biologists and biophysicists use it to train and benchmark models that evaluate binding free energy changes, interface destabilisation, and protein complex stability.
How it was made
The dataset compiles 7,085 curated mutation entries linked to specific PDB structures and protein chains. Each entry pairs wild-type and mutant dissociation constants in molar units with experimental metadata, such as assay temperature, measurement method, and source literature identifiers. Where reported in the original studies, kinetic association rates (kon), dissociation rates (koff), enthalpy changes (dH), and entropy changes (dS) are also recorded.
In this distribution, the raw upstream semicolon-separated CSV records are provided as JSONL entries with row indices and provenance tracking. Rows contain both original and cleaned mutation notations alongside structural classifications designating whether a mutation occurs in the core, rim, support, or surface of the interaction interface.
Key facts
- Total records
- 7,085 mutation entries
- Primary format
- JSONL
- Dataset splits
- Single split (train: 7,085 rows)
- Primary measurement
- Binding affinity dissociation constant (Kd in M)
- Secondary measurements
- kon, koff, dH, and dS
- Structural annotation
- Protein Data Bank (PDB) complexes, chains, and interface location classes
What you can use it for
- Predicting binding free energy changes
- Benchmark machine learning and physics-based models that predict changes in binding free energy (ddG) from wild-type and mutant dissociation constants.
- Interface mutation profiling
- Assess how point and multiple mutations affect protein-protein interfaces across distinct spatial categories, such as core, rim, and surface regions.
- Kinetic and thermodynamic modelling
- Evaluate predictive tools on experimental association and dissociation rates or enthalpy and entropy shifts when kinetic data are available.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from LiteFold/SKEMPI2 at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 7,085 rows of default/train
dataset_id
string_label- skempi2100%
row_index
int- min
- 0
- median
- 3,542
- mean
- 3,542
- max
- 7,084
source_file
string_label- labeled/skempi2/skempi_v2.csv100%
Strengths and limits
Good at
- Directly pairs experimental binding and kinetic measurements with structurally resolved PDB complexes.
- Includes thermodynamic enthalpy and entropy values alongside association and dissociation rates.
- Annotates interface locations into structural categories including core, rim, support, and surface.
Watch out for
- Does not supply predefined train and test splits, requiring users to design grouping splits to avoid data leakage across shared complexes or protein families.
- The JSONL mirror stores upstream CSV values under a semicolon-delimited header key, requiring extra parsing before tabular use.
- Some entries lack explicit experimental temperatures or contain multi-mutation records and unparsed values that require filtering.
Questions
Can I look inside SKEMPI2 without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is SKEMPI2 under?
Its authors declare other. Read the terms on the original page before reusing the data.
What units are used for binding affinities?
Affinities for mutant and wild-type proteins are reported as dissociation constants in molar (M) units.
How is binding free energy change calculated from these entries?
Binding free energy change is typically computed as ddG = R * T * ln(Kd_mut / Kd_wt), where R is the gas constant, T is the experimental temperature in Kelvin, and Kd represents the parsed dissociation constants.
Why is random cross-validation discouraged on this dataset?
Multiple records frequently share the same PDB complex, protein pair, or related mutations. Random splitting risks data leakage; evaluations should instead hold out entire complexes, protein pairs, or interaction families.
What structural interface categories are provided?
Mutation sites are classified into interface structural regions such as core, rim, support, or surface.
How to cite it
Jankauskaitė, J., Jiménez-García, B., Dapkūnas, J., Fernández-Recio, J., Moal, I. H. (2019). SKEMPI 2.0: an updated benchmark of changes in protein-protein binding energy, kinetics and thermodynamics upon mutation. Bioinformatics, 35(3), 462–469. https://doi.org/10.1093/bioinformatics/bty635
Written from SKEMPI2's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.