Work

Support Vector Machine Screening for Dual Adenosine A2A Receptor Antagonists and Monoamine Oxidase B Inhibitors

Year: 2026
Type: article
Abstract: Background: Adenosine A2A receptor antagonism and monoamine oxidase B (MAO-B) inhibition offer complementary approaches to motor fluctuations in Parkinson disease. This historical study examined ligand-based classifiers for prioritizing compounds predicted to have both activities. Methods: The 2011 analysis used separate support vector machine (SVM) models with 98 molecular descriptors. Training sets contained 2,169 reported A2A receptor antagonists and 63,118 presumed inactive compounds, and 187 MAO-B inhibitors and 66,366 presumed inactive compounds. The reported activity threshold was a half-maximal inhibitory concentration (IC50) or inhibition constant (Ki) below 10 micromolar. Five-fold cross-validation was used for parameter selection. Evaluation included later-reported active-compound sets and a separate 14-compound dual-target benchmark. Screening involved approximately 13.56 million PubChem entries and 168,000 Molecular Design Limited Drug Data Report (MDDR) entries. This revision reports surviving aggregate results; no new screen was performed. Results: The A2A receptor model retrieved 210 of 315 later-reported antagonists (66.7%), and the MAO-B model retrieved 84 of 88 inhibitors (95.5%). The receptor classifier retrieved nine of 14 dual-target benchmark compounds (64.3%); both classifiers jointly retrieved five (35.7%). Reported single-target positive counts were 8,285 and 1,307 in PubChem and 1,507 and 69 in MDDR for the receptor and enzyme models, respectively. The historical intersection count was 58. Candidate identities and individual outputs are unavailable, preventing independent verification and confirmation of uniqueness. Training/evaluation structural overlap could not be audited, and unreconciled MAO-B internal metrics were excluded from quantitative comparison. Conclusions: Models trained on pre-2010 data retrieved most compounds in the selected later-reported single-target active sets, while joint retrieval was lower in the separate dual-target benchmark. These findings describe a historical screening approach but do not establish prospective accuracy or useful dual pharmacology. Missing candidate identities and computational records prevent independent verification of the 58-entry screening intersection and limit its use for experimental follow-up.
Source: Cureus
Author: Ikken Aisin
Language: English
Subfield: Drug Discovery
Open Access status: diamond
OpenAlex ID: W7236317362
FWCI:
Cites: 24
Cited by:
Related to: 10