PPI Disruptors Are Not Typical Drug Candidates
The pharmacokinetic and toxicity challenges facing PPI-targeting small molecules are well-documented in the medicinal chemistry literature, and they are systematically different from those of enzyme inhibitors. Understanding this is essential for anyone applying standard ADMET filters — Lipinski's Rule of Five, Veber's rules for oral bioavailability, standard hERG and CYP liability cutoffs — to a PPI disruptor candidate set. The mismatch between standard filters and PPI disruptor chemical space is not a minor nuance; it is a significant source of false rejection that has demonstrably affected PPI program outcomes.
The root cause is structural. PPI disruptors need molecular surface area sufficient to span a hot-spot sub-pocket geometry that is larger and more exposed than typical enzyme active sites. This requirement drives compound characteristics upward along several ADMET-relevant axes simultaneously: molecular weight, lipophilicity, molecular complexity, and often molecular flexibility. The result is a class of compounds that sits outside the Ro5 space where the most favorable drug-like ADMET profiles are concentrated.
The field has developed the "beyond Rule of Five" (bRo5) concept to address this, and the most successful PPI-targeting drugs on the market — venetoclax being the clearest example — demonstrate that compounds outside the classical Ro5 space can achieve acceptable oral bioavailability and therapeutic windows, but the pathway to achieving those properties requires different medicinal chemistry strategies and different ADMET prediction tools than Ro5-space programs use.
Lipophilicity: The Double-Edged Property
Lipophilicity (measured as cLogP or experimentally as LogD at pH 7.4) is the single most important ADMET-relevant property to get right for PPI disruptors, and it is the one that standard filters handle most poorly. The typical Lipinski cutoff of cLogP <5 is calibrated to prevent compounds from being too lipophilic for acceptable aqueous solubility and metabolic stability. But PPI disruptors need hydrophobicity to displace water from hot-spot sub-pockets and make productive hydrophobic contacts with non-polar interface residues. The "correct" lipophilicity range for PPI disruptors is systematically higher than for kinase inhibitors — cLogP 3–6 is a reasonable starting range, and many successful PPI disruptors sit at the upper end of this range or slightly above it.
The ADMET consequences of higher lipophilicity are predictable: increased plasma protein binding (reducing free fraction), higher metabolic clearance by CYP enzymes (increased hepatic clearance), reduced aqueous solubility (affecting dissolution and bioavailability), and increased risk of non-specific binding artifacts in biophysical assays (a practical problem in hit identification). None of these consequences are disqualifying on their own — they are design problems to be solved, not filters to be applied. A compound with cLogP 5.5 that has been appropriately optimized for solubility (by adding a solubilizing group without increasing MW excessively), metabolic stability (by blocking metabolically labile positions identified by microsomal incubation), and plasma protein binding (by calibrating against total and free concentration data) is a viable lead. The same compound that has not been optimized for these properties is not — but the appropriate response is optimization, not rejection at the screening stage.
Standard computational ADMET tools trained primarily on Ro5-space compounds will systematically underestimate the achievable ADMET performance in the bRo5 chemical space, because their training sets are biased. There is an emerging body of bRo5-specific ADMET data — driven in part by the success of macrocyclic drugs, PROTACs, and large natural product derivatives — that is beginning to improve the calibration of computational ADMET models in this space, but most commercial tools still underperform for PPI disruptor molecular weight ranges.
Molecular Weight and Its Consequences
PPI-targeting lead compounds typically fall in the MW 450–700 Da range. This creates several ADMET challenges that are mechanistically distinct from the lipophilicity issue. Passive membrane permeability decreases with MW above approximately 450 Da, following well-characterized size-dependent permeability models. Oral bioavailability correlates negatively with MW in this range even when LogD is held constant, because the enthalpy of desolvation and the entropic cost of conformational restriction in the lipid bilayer both scale unfavorably with molecular size.
The medicinal chemistry response to this challenge in the context of PPI disruptors has typically been one of two strategies. The first is deliberate molecular constraint: introducing macrocyclization or conformational rigidification to reduce the conformational entropy penalty of membrane permeation. Constraining a flexible PPI disruptor into a more rigid shape reduces the conformational entropy cost of achieving the bioactive conformation, and for macrocyclic compounds, can also reduce the desolvation cost of membrane permeation by pre-organizing the polar groups into an inward-facing arrangement. The second strategy is to accept reduced passive permeability and compensate with active transport, particularly for cancer indications where target tissue access may involve mechanisms beyond passive diffusion (tumor-selective uptake, lysosomal trapping, or direct injection for localized delivery).
Computational prediction of membrane permeability for bRo5 compounds requires models calibrated on bRo5 permeability data — specifically Caco-2 or PAMPA measurements from compounds in the MW 450–700 range. The commonly used permeability prediction models (trained primarily on MW 200–450 compounds) apply poorly in this range. At Genolux, we apply MW-range-specific permeability models in our ADMET prediction pipeline rather than using general-purpose tools uncalibrated for high-MW compounds.
hERG Liability Assessment for High-MW Lipophilic Compounds
hERG channel blockade — the primary cause of drug-induced QT prolongation — is a function of molecular charge state, lipophilicity, and the presence of specific pharmacophore features that interact with the hERG channel cavity. The canonical hERG pharmacophore involves a basic nitrogen and a hydrophobic region at a specific spatial relationship, but the liability is not restricted to this pattern. High lipophilicity alone increases hERG risk, and PPI disruptors' tendency toward higher cLogP means hERG liability is a real concern that cannot be assessed with the rapid qualitative filters appropriate for Ro5-space compounds.
For PPI disruptor series, we run hERG predictions using ensemble models that include both pharmacophore-based and ML-based QSAR models trained on patch-clamp hERG IC₅₀ data. The pharmacophore model is more interpretable (flagging specific chemical features associated with hERG risk) while the ML model has better global coverage of chemical diversity. We use both and flag compounds where either model predicts IC₅₀ <10 μM for hERG blockade. For compounds flagged by hERG predictions, we annotate the structural features responsible so that the medicinal chemistry team has specific modification suggestions rather than a binary pass/fail output.
Adapted ADMET Workflows for PPI Discovery
The practical adjustment for PPI programs is not to abandon standard ADMET assessment — it is to use assessment tools calibrated for bRo5 chemical space and to apply compound-specific rather than class-generic thresholds. Venetoclax has a MW of 868 Da and cLogP approximately 7.3 — it would be rejected at the first filter stage of any standard ADMET screening protocol. It is orally bioavailable in the clinical setting because its medicinal chemistry development understood the specific structural modifications needed to achieve acceptable oral ADMET despite those physicochemical flags.
We're not suggesting that MW 700+ and cLogP 7+ are acceptable starting points for every PPI disruptor program — the ADMET challenges compound at extreme values, and the medicinal chemistry investment to address them is substantial. What we are saying is that the typical Ro5/Lipinski filter applied as a hard cutoff during computational virtual screening will reject a substantial fraction of the chemical space that is specifically appropriate for PPI interface engagement. The correct approach is to apply adapted PPI-specific ADMET filters that reflect the physicochemical requirements of the target class, flag the property risks clearly for medicinal chemistry attention, and advance the compounds with the best disruption potential alongside an explicit ADMET optimization plan — not to pre-filter the PPI-appropriate chemical space out of existence before the compound has been synthesized.
The distinction between a compound that is outside Ro5 because of non-productive promiscuity and one that is outside Ro5 because the target requires it is a critical one to make correctly. Standard ADMET filtering cannot make this distinction — it requires target-class context. For PPI programs, applying that context is the difference between a productive hit identification campaign and one that systematically rejects its best candidates before they reach synthesis.