Plasma Protein Binding Data for 1651 Chemicals with QSAR Predictions

In vitro plasma protein binding values for 1245 pharmaceuticals and 406 ToxCast chemicals were collected from literature published between 2008 and 2015. The dataset contains 21 molecular descriptors calculated by MOE, an acid/base/neutral/zwitterion classification, and in silico predictions from three QSAR models (k-nearest neighbor, support vector machine, random forest) plus a consensus prediction. It was compiled by the U.S. Environmental Protection Agency to support research on chemical binding predictability.

Use Cases

• Compare the k-nearest neighbor, support vector machine, and random forest QSAR model predictions for fraction unbound (Fub) across the 1651 chemicals.
• Analyze the relationship between the 21 MOE-calculated molecular descriptors and the measured in vitro plasma protein binding values.
• Investigate how the acid/base/neutral/zwitterion classification correlates with model prediction accuracy or consensus prediction values.
• Validate the domain of applicability for QSAR models by examining prediction consistency for pharmaceuticals versus ToxCast chemicals.

Strengths

• Contains experimental binding data for a substantial set of 1651 chemicals (1245 pharmaceuticals + 406 ToxCast chemicals).
• Includes predictions from three distinct machine learning models (k-NN, SVM, RF) plus a consensus, enabling comparative analysis.
• Provides 21 calculated molecular descriptors and a chemical ionization classification, offering multiple analytical features.

Limitations

• The underlying experimental data is compiled from literature sources spanning 2008-2015, which may not reflect the most recent compounds or measurement techniques.
• The dataset's applicability is focused on environmentally relevant chemicals and pharmaceuticals, limiting generalizability to other chemical classes.
• The exact number of rows and the structure of the data file (XLS) are unspecified, which may complicate programmatic access.

Provenance

Source

U.S. Environmental Protection Agency.

Collection Method

In vitro values collected from published literature (Obach 2008, Zhu 2013, Wetmore 2012, Wetmore 2015); descriptors calculated by MOE; predictions generated by constructed QSAR models.

Time Range

Literature sources span from 2008 to 2015.

Freshness

Last updated in November 2020.