Membrane Access and Orbital Localization Govern ABC Transporter Substrate Recognition

The ATP-binding cassette transport protein Pdr5p is known to play a role in multidrug resistance in Saccharomyces cerevisiae by effluxing structurally diverse xenobiotics; yet the physicochemical determinants of substrate recognition remain poorly defined. To address this, density functional theory (DFT) calculations at the B3LYP-D3BJ/def2-SVP level were combined with machine learning to derive a predictive model of substrate recognition using a curated dataset of 66 compounds spanning 9 functional categories. A hybrid support vector machine (SVM) classifier achieved 96.3% accuracy (95% CI: 81.0-99.9%, Clopper-Pearson exact) in discriminating substrates from non-substrates under leave-one-out cross-validation. Feature importance analysis identified lipophilicity (LogP, F-score = 37.5) as the dominant descriptor, suggesting that membrane partitioning constitutes the initial recognition step. The HOMO-LUMO gap contributed secondarily (F-score = 12.4). Substrates were further distinguished by high frontier orbital focalization, with frontier orbital spread of 1.8-2.6%, compared to 4.18-7.22% for non-substrates. Notably, a model trained exclusively on Pdr5p data achieved 87% prediction accuracy when applied without retraining to the human P-glycoprotein (ABCB1) dataset, suggesting conserved physicochemical principles of substrate recognition across evolutionarily distant ABC transporters. These findings provide a quantum chemical framework for understanding and potentially predicting MDR transporter substrate specificity.