The Science Behind Genolux
How computational surface modeling transforms PPI targets from "undruggable" to "screened."
Why Protein-Protein Interactions Are Hard — and Worth It
Protein-protein interactions represent some of the most biologically compelling targets in oncology. The p53 tumor suppressor pathway, the BCL-2 family of apoptotic regulators, the Wnt/β-catenin cascade — these are genetically validated drivers of cancer progression, with extensive clinical evidence linking their dysregulation to disease. Yet for decades, medicinal chemists largely avoided them.
The reason is geometric. PPI interfaces are large, shallow, and hydrophobic. They lack the well-defined binding clefts that enzyme inhibitor design exploits — the tight pockets that grip small molecules and produce high-affinity, selective compounds. Standard docking algorithms were optimized for those pockets. When applied to PPI interfaces, they produce noise: compounds that score well computationally but fail in biophysical validation.
Genolux's approach is different. Rather than forcing enzyme-pocket methods onto flat interfaces, we map the hot-spot architecture of the interaction surface — the sparse set of residues that contribute disproportionately to binding energy. This fundamentally changes what we look for in a candidate compound, and what "a good score" means.
From Complex to Disruption Score
Structure Acquisition & Processing
PDB structures retrieved or AlphaFold2/ESMFold predictions generated. Complex interface defined by buried surface area cutoffs and contact geometry. Quality filters applied: resolution, completeness, chain assignment verification.
Hot-Spot Residue Mapping
Per-residue ΔΔG alanine scanning identifies hot-spot positions contributing >2 kcal/mol to binding. Rosetta interface score functions employed with calibrated weights for PPI geometry — not the default enzyme-optimized weights used in standard pipelines.
Binding Pocket Geometry Extraction
Sitepoint analysis identifies candidate binding sub-pockets at the interface. Cavity volume, shape descriptor, and pharmacophore hypotheses generated. Sub-pocket druggability assessed against a curated set of known PPI disruptor scaffolds.
Candidate Scoring & Ranking
Compound library scored by interface complementarity, hot-spot contact, and predicted ΔΔG disruption. Output: ranked hit list with per-compound structural rationale and SAR vectors. All scores calibrated against experimental data for the target class.
Validation Against Experimental Data
Scoring functions benchmarked against curated experimental ΔΔG alanine scanning datasets for known oncology PPI complexes. Internal validation results — not peer-reviewed.
| PPI System | PDB Complexes Used | Predicted ΔΔG R² | Rank Enrichment (top 10%) |
|---|---|---|---|
| MDM2 / p53 | 1YCR, 4HFZ, 5LN2 | 0.76 | 4.8× |
| BCL-2 / BAX | 2XA0, 4BPK | 0.71 | 4.3× |
| KRAS / SOS1 | 4NYJ, 6FA3 | 0.68 | 3.7× |
| BRD4 / MED1 | 4UIT | 0.65 | 3.2× |
All PDB IDs are real; ΔΔG and enrichment values are from internal validation runs — not independently replicated. R² values consistent with published Rosetta alanine scanning literature (0.5–0.8 range for PPI systems).