ERpred

ERpred predicts subtype-specific antagonists targeting estrogen receptors alpha (ERα) and beta (ERβ) to support identification of selective estrogen receptor modulators.

Key Features:

• CSAR data source: Built from a classification structure-activity relationship (CSAR) study using ChEMBL datasets initially comprising 11,618 compounds for ERα and 7,810 compounds for ERβ.
• Bioactivity unit and curation: IC50 was used as the bioactivity unit and curation yielded final datasets of 1,593 compounds for ERα and 1,281 compounds for ERβ.
• Machine learning algorithm: Random forest (RF) models were trained for subtype-specific classification.
• Molecular descriptors: Multiple fingerprint types were evaluated and the PubChem fingerprint demonstrated superior robustness.
• Performance metrics: PubChem-fingerprint RF models achieved accuracies of 94.65% for ERα and 92.25% for ERβ, with Matthews correlation coefficients reported as 89% for ERα and 76% for ERβ.
• Feature interpretation: Feature importance analysis highlighted aromatic rings, nitrogen-containing functional groups, and aliphatic hydrocarbons as influential for bioactivity.

Scientific Applications:

• Subtype-specific antagonist prediction: Predicts compound bioactivity specifically against ERα and ERβ.
• Selective modulator design: Guides design of selective estrogen receptor modulators (SERMs) with ER subtype specificity.
• Oncology drug discovery: Supports identification of compounds relevant to breast and endometrial cancer research related to ER signaling.
• Adverse-effect assessment: Assists evaluation of subtype specificity to mitigate risks associated with estradiol and hormone replacement therapies, including thromboembolic and cancer-related risks.

Methodology:

Conducted a CSAR study using ChEMBL datasets (initially 11,618 ERα and 7,810 ERβ compounds) with IC50 as the bioactivity unit, curated to final sets (1,593 ERα; 1,281 ERβ), trained random forest models on molecular fingerprints (PubChem fingerprint selected), evaluated performance by accuracy and MCC, and performed feature importance analysis identifying key chemical moieties.