Conclusions

The present study introduces a chromatography-informed modelling framework for predicting BSEP inhibition using experimentally derived descriptors of membrane affinity and protein-associated distribution. The results show that biomimetic chromatographic descriptors, particularly CHI IAM and log _k_HSA, provide an interpretable and experimentally accessible way to capture distribution-related determinants of BSEP inhibition that are not fully represented by conventional physicochemical parameters such as log P and log _D_7.4.

A central finding of this work is that membrane affinity, represented by CHI IAM, is a dominant determinant of BSEP inhibition in the present dataset. The IAM-only model showed strong predictive performance and external predictivity comparable to the two-descriptor biomimetic model, supporting its potential use as a practical first-tier screening approach when experimental simplicity and throughput are priorities. The addition of log _k_HSA modestly improved training and internal cross-validation performance and provided complementary mechanistic information on albumin-binding propensity, highlighting a practical trade-off between experimental simplicity and broader biomimetic characterization.

The improved performance of biomimetic models relative to conventional log P/log D-based models supports the hypothesis that BSEP inhibitory liability is not adequately described by bulk lipophilicity alone but is also influenced by distribution-related factors that determine compound enrichment and interaction at the membrane interface. This is particularly relevant for BSEP, which is localized at the canalicular membrane, where membrane-proximal compound enrichment and protein-associated distribution may influence the effective concentration available for transporter interaction. Conventional lipophilicity descriptors remain useful reference parameters, but they cannot fully account for ionization-dependent membrane interactions, phospholipid affinity, or protein binding.

Importantly, the proposed model should not be interpreted as a direct predictor of clinical DILI. Clinical DILI is a multifactorial outcome influenced by metabolic processes, mitochondrial toxicity, immune-mediated mechanisms, dose and exposure, and patient-specific susceptibility. Instead, the present approach is best positioned as a mechanistically informed screening layer for identifying compounds with potential BSEP-mediated cholestatic liability. It can complement, but not replace, in vitro transporter assays, hepatocyte-based systems, and integrated DILI risk-assessment frameworks.

This study demonstrates the value of biomimetic chromatography as a bridge between physicochemical characterization and mechanistic toxicology. By linking experimentally measured membrane and protein-binding interactions to transporter inhibition, the proposed framework provides a practical, interpretable, and scalable tool for early compound prioritization. Future work should focus on expanding the dataset's chemical space, validating the approach across larger external compound sets, incorporating additional mechanistic descriptors related to intracellular accumulation and metabolism, and integrating biomimetic models into multi-parameter DILI risk-assessment strategies.