metadata

license: mit
task_categories:
  - feature-extraction
language:
  - en
tags:
  - biology
pretty_name: 'ESMnrg: Protein Stability Prediction Dataset'
size_categories:
  - 100K<n<1M

Protein Stability Prediction Dataset (PSPD)

The Protein Stability Prediction Dataset (PSPD) is a curated collection of protein sequences and their corresponding stability measurements, specifically the Gibbs free energy changes (ΔG) upon mutation. This dataset is designed to facilitate the development and evaluation of computational models for predicting the impact of mutations on protein stability, sourced from existing literature sources.

Dataset Description

The PSPD dataset contains the following key information:

Amino acid sequences of proteins
Gibbs free energy changes (ΔG) associated with mutations
Binned ΔG values for classification tasks

The dataset is curated from multiple protein stability data sources in the existing literature, ensuring a diverse and representative collection of proteins and mutations.

Data Format

The dataset is provided in a CSV format, with each row representing a protein entry. The columns include:

aa_seq: Amino acid sequence of the protein
deltaG: Gibbs free energy change (ΔG) upon mutation
deltaG_bin: Binned ΔG values for classification tasks

Example entry: aa_seq,deltaG,deltaG_bin MKIFVKTLTGKTITLEVEPSDTIENVKAKIQDEEGIPPDQQRLIFAGKKLEDGRTLTDYSIQKESTLHLVLR,8.09999848220399,8 to 9

Data Preparation

The dataset undergoes several preprocessing steps to ensure data quality and consistency:

Cleaning: Null, duplicate, incomplete, or incorrect entries are removed.
Extraction: Relevant information (amino acid sequences, ΔG values, binned ΔG values) is extracted.
Normalization: Quantitative inputs are verified and scaled to conform to standard units and scales.
Augmentation: The dataset is augmented to address the imbalanced distribution of energy values.

Applications

The PSPD dataset is suitable for various applications in computational biology and protein engineering, including:

Developing and benchmarking machine learning models for predicting protein stability changes upon mutation
Investigating the relationship between protein sequence and stability
Guiding rational protein design and mutagenesis studies

License

The PSPD dataset is released under the MIT license. By using this dataset, you agree to abide by the terms and conditions of the license.