The Abundance of Bad PPI Targets
The number of protein-protein interactions in the human interactome that are implicated in cancer biology is large — estimates based on large-scale protein interaction network analyses and cancer genomics data suggest thousands of PPIs have some degree of evidence for relevance to oncogenesis. The number of those PPIs that represent viable drug discovery targets is substantially smaller. The gap between "biologically implicated" and "therapeutically tractable as a PPI disruption target" is the central challenge in PPI program selection, and the criteria for distinguishing tractable from intractable targets are worth making explicit.
Early PPI drug discovery programs that have failed have often failed not because the biology was wrong — the cancer relevance of the target was real — but because the target selection process did not adequately weigh the structural and chemical tractability of the interface against the biological validation. The result was investing substantial medicinal chemistry resources against targets where the interface geometry was fundamentally incompatible with small-molecule disruption, or where disruption of the interface in cancer cells would also disrupt essential normal cellular functions, or where the Kd of the native protein-protein interaction was so tight that the compound concentration required for competitive disruption was incompatible with achievable in vivo exposure.
A structured target selection framework prevents these failures, or at least surfaces the key risks early enough to allow an informed decision about whether to proceed. The framework we use at Genolux evaluates targets across three domains: biological validation quality, interface structural tractability, and chemical tractability. All three need to pass a threshold for a target to warrant substantial resource investment.
Domain 1: Biological Validation Quality
Biological validation for a PPI target has several components that need to be distinguished. The most basic question is whether disrupting this specific PPI interface has been shown to produce an anti-tumor effect in a relevant model. "Relevant model" is important: a target validated only in cell-free biochemical assays or in a highly artificial cellular system is at a lower confidence level than one where disruption of the PPI — using a genetic approach, a peptide mimetic, or a validated small molecule — has been shown to produce anti-proliferative or pro-apoptotic effects in cancer cell lines with documented genotype information.
Beyond in-cell validation, genetic evidence from human cancer genomics strengthens the case significantly. A PPI target where one or both interacting proteins shows gain-of-function mutation, amplification, or overexpression in the cancer types being targeted, or where loss of a negative regulatory PPI is a documented resistance mechanism, has a biological rationale that is grounded in human cancer biology rather than extrapolated from model systems. MDM2 overexpression in TP53-wildtype tumors is the classic example: the genetics directly implicates the MDM2-p53 PPI in tumor biology in human patients.
One important qualifier: genetic validation of the protein does not automatically validate the specific PPI interface as the therapeutic target. MDM2 has multiple interaction partners beyond p53 — disrupting the MDM2-p53 interface is specifically validated as the mechanism, but disrupting other MDM2 interactions might not produce the desired anti-tumor effect. The biological validation should specifically support the PPI being disrupted, not just the protein family.
Domain 2: Interface Structural Tractability
Structural tractability assessment begins with the questions discussed in detail throughout this blog series: interface area, hot-spot distribution, sub-pocket geometry, interface dynamics, and available structural data quality. Several quantitative thresholds guide this assessment, though these should be treated as soft filters rather than hard cutoffs.
Interface area in the 800–2,000 Ų range is generally more tractable than either smaller or larger interfaces. Very small interfaces (<600 Ų) may lack the hot-spot concentration needed for a small molecule to achieve functional disruption — a single residue at a 400 Ų interface is not a hot spot in the same sense as a residue at a 1,200 Ų interface where that residue contributes 30% of the total binding energy. Very large interfaces (>2,500 Ų) impose large MW requirements on disruptors that challenge bioavailability.
Hot-spot concentration is probably the more important filter. An interface with 3–5 hot-spot residues contributing >2 kcal/mol each, concentrated in a geometrically defined sub-pocket of volume 200–500 ų, is significantly more tractable than one where the binding energy is distributed across 20 residues with no single large contributor. The MDM2-p53 interface is tractable in part because F19 and W23 together account for a large fraction of the binding ΔG, and they insert into a well-defined hydrophobic cleft that is geometrically suited to small-molecule mimicry.
Interface dynamics — characterized by MD as discussed previously — needs to show that the hot-spot sub-pocket is accessible with reasonable frequency. A cryptic pocket that opens only 5% of the MD simulation time is not a reliable small-molecule binding site for a drug discovery program, even if it is geometrically favorable when open. Binding frequency thresholds in the 20–30% range are more appropriate for drug discovery confidence.
Domain 3: Chemical Tractability and Drug-Like Space Availability
Even a biologically validated, structurally tractable PPI interface may be chemically intractable if the compound properties required for effective disruption are incompatible with acceptable oral ADMET profiles. The assessment here requires estimating the MW and lipophilicity range needed to achieve hot-spot engagement, and evaluating whether compounds in that range can realistically achieve the pharmacokinetic exposure necessary for in vivo efficacy.
This is where the bRo5 (beyond Rule of Five) analysis becomes critical for PPI targets. A target where the sub-pocket geometry demands compounds with MW >600 Da to simultaneously engage F19-, W23-, and L26-equivalent hot spots is chemically tractable only if the team has the medicinal chemistry capability to work in bRo5 space — which requires different library design, different synthetic strategies, and different ADMET optimization approaches than typical kinase inhibitor programs. For organizations without this capability, a target requiring bRo5 compounds is a higher-risk selection than one where the sub-pocket can be addressed by lead-like compounds in the MW 400–500 range.
Chemical tractability also encompasses the competitive landscape: if multiple known compounds already demonstrate that the target is druggable — as is the case for MDM2-p53 where multiple advanced small-molecule disruptors have been characterized crystallographically and studied in cellular models — the chemical starting point risk is substantially lower. A target with no known small-molecule binders requires demonstration of chemical tractability as a first step, which is appropriately treated as a risk item in the target selection decision.
The Tradeoff Triangle and How to Use It
The three domains — biological validation, structural tractability, and chemical tractability — define a tradeoff triangle. Most targets are strong in one or two domains and weaker in the third. The target selection decision is about whether the strength in the favorable domains is sufficient to justify the investment required to address the weakness in the unfavorable domain.
A target with strong biological validation and favorable structural tractability but uncertain chemical tractability (no known binders) warrants a focused fragment-based exploration to determine chemical tractability before committing to a full lead optimization campaign. A target with strong biological validation and known chemical tractability but a computationally complex interface (large, flat, diffuse hot spots) warrants significant upfront structural characterization investment, including experimental alanine scanning if resources allow, to identify whether there are sub-pockets not apparent from the initial computational analysis before committing to synthesis. A target with high structural tractability but weak biological validation — perhaps a computationally beautiful interface but only cell-free validation of the PPI's oncological relevance — warrants investment in cellular and, ideally, genetic validation before structural work becomes a major resource item.
Targets We Have Deprioritized and Why
We're not saying every computationally tractable PPI interface in oncology is worth pursuing — the field has learned hard lessons from programs that had excellent chemistry but insufficient biological validation, or programs where structural tractability did not translate to cellular engagement because of compound permeability limitations. Our own target prioritization has led us to deprioritize several initially attractive interfaces: one case involved an interface where MD analysis revealed the hot-spot sub-pocket is less stable than the crystal structure suggested, with pocket volume varying over a factor of three across a 500 ns trajectory; another involved a target where the biological validation relied heavily on genetic knockdown data that poorly models the pharmacological consequence of competitive PPI disruption (genetic deletion of a protein eliminates all its functions, not just the specific PPI being targeted).
Target selection is ultimately a resource allocation decision under uncertainty. The framework described here is not a guarantee of success — it is a way of making the uncertainty explicit, assigning the risks to the right domains, and concentrating resources on targets where multiple lines of evidence converge. The best PPI targets have genetic validation from human cancer data, structural tractability confirmed computationally and ideally experimentally, and chemical tractability supported by at least one known small-molecule binding series as a starting point. When all three align, the remaining risk is in the execution — the chemistry, the ADMET optimization, and the translational biology. When they don't all align, the early-stage question is which gap is the most important to address before the program moves forward.