Machine learning models based on residue interaction network for ABCG2 transportable compounds recognition.

Area of Science:

• Biochemistry
• Pharmacology
• Toxicology

Background:

• The ATP-binding cassette sub-family G member 2 (ABCG2) is vital for transporting environmental substances across the placenta.
• Accurate identification of ABCG2 transport molecules is essential for evaluating placental exposure risks to environmental chemicals.

Purpose of the Study:

• To explore differences in ABCG2 binding conformations between transportable and non-transportable compounds.
• To develop a predictive model for ABCG2 molecular recognition using quantitative residue interaction network (RIN) data.

Main Methods:

• Residue interaction network (RIN) analysis was employed to quantitatively characterize protein conformations.
• Four machine learning algorithms were applied to build classification and recognition models for 1100 compounds.
• Random Forest (RF) models were specifically evaluated for their predictive performance.

Main Results:

• Quantitative RIN data effectively captured changes in amino acid conformation and inter-amino acid interactions.
• The developed machine learning models demonstrated strong classification and recognition capabilities.
• Random Forest models achieved high predictive accuracy, with 0.97 for the training set and 0.96 for the test set.

Conclusions:

• Constructing residue interaction networks offers a novel approach for quantitative protein conformation characterization.
• The ABCG2 molecular recognition model based on RIN provides a valuable tool for screening environmental chemicals that cross the placenta.
• This methodology aids in assessing potential risks associated with placental transport of environmental compounds.