Biomimetic chromatographic descriptors for modeling therapeutic, toxic, and lethal blood concentration endpoints:...

Blood concentration thresholds associated with therapeutic, toxic, and lethal effects reflect distribution-related processes but also pharmacodynamic, clinical, and forensic factors. This study examined whether biomimetic chromatographic descriptors provide information distinct from conventional physicochemical descriptors, or primarily serve as integrative proxies for the same underlying descriptor space. A curated dataset of 92 drugs was modeled using biomimetic descriptors, including CHI_IAM, log k_HSA, and log k_AGP, alongside conventional variables such as log P, log D₇.₄, molecular weight, hydrogen-bonding, polarity, and ionization descriptors. CHI_IAM was the most informative single descriptor across all endpoints (R² ≈ 0.25-0.32), indicating that membrane-affinity-related properties contribute to systemic concentration thresholds. Multivariate models improved performance substantially (R² ≈ 0.62-0.73; Q²_LOO ≈ 0.57-0.70), but biomimetic and conventional models showed nearly equivalent resampling performance (Q²_pred ≈ 0.44-0.59). Nested model analysis showed that adding CHI_IAM improved incomplete conventional models (ΔR² up to 0.28, p < 0.001), whereas adding conventional descriptors to CHI_IAM-based models provided limited additional benefit. These findings indicate that CHI_IAM functions as a compact experimental descriptor integrating lipophilicity, ionization, polarity, and molecular-size effects that conventional models reconstruct through multiple calculated variables. Robustness was supported by repeated validation, Y-randomization, and applicability-domain analysis, although predictive performance remained moderate and internally validated. Biomimetic chromatography offers an experimentally grounded low-dimensional representation of distribution-related physicochemical space, whereas conventional descriptors provide greater mechanistic decomposability. These results support the flexible integration of biomimetic and calculated descriptors for early-stage drug distribution and toxicological modeling, while emphasizing their use for descriptor interpretation and comparative screening rather than for direct clinical or forensic prediction.