QEPPI

Quantitative estimate for protein-protein interaction inhibitor potential. Uses QEPPI algorithm to evaluate drug-likeness for compounds targeting PPIs.

Input

Job name

Molecule

Output

Configure input settings on the left, then click "Submit"

What is QEPPI?

QEPPI (Quantitative Estimate Index for Compounds Targeting Protein-Protein Interactions) scores how drug-like a molecule is when targeting protein-protein interactions. Traditional drug-likeness rules like Lipinski's Rule of 5 were designed for enzyme inhibitors—they fail for PPI inhibitors, which need different properties.

The problem: PPI binding surfaces are large and flat, requiring bigger, more lipophilic molecules to disrupt them. Applying Lipinski's rules would reject most successful PPI drugs. QEPPI solves this by scoring molecules against 1,007 experimentally validated PPI inhibitors from the iPPI-DB database rather than conventional drugs.

Use QEPPI early in drug discovery to filter virtual libraries, prioritize compounds for synthesis, or guide medicinal chemistry optimization toward PPI-favorable chemical space.

When to use QEPPI vs alternatives

Use QEPPI when:

Targeting protein-protein interactions specifically
Screening large compound libraries (>10,000 molecules)
Optimizing lead compounds for PPI targets like Bcl-2, MDM2, or bromodomains
You need continuous scoring (0-1) rather than binary pass/fail

Use Lipinski's Rule of 5 when:

Targeting traditional enzyme active sites or receptors
Designing conventional oral drugs
You want simple binary filtering (pass/fail)

Use ADMET-AI when:

You need absorption, distribution, metabolism, excretion predictions
Assessing toxicity risks beyond drug-likeness
Later-stage lead optimization (QEPPI is for early screening)

Use Molecular Descriptors when:

You want raw physicochemical properties without scoring
Building custom scoring functions or QSAR models

How QEPPI works

QEPPI calculates seven molecular properties from SMILES strings, converts each to a "desirability score" using sigmoid curves fitted to successful PPI drugs, then combines them into a single score from 0 to 1.

Molecular weight (MW): PPI inhibitors peak around 500 Da versus 350 Da for conventional drugs. The algorithm accepts molecules up to 600 Da, reflecting the larger binding surfaces that must be covered.
Lipophilicity (ALogP): Optimal at 4.8 versus 2.7 for traditional drugs. PPI interfaces are often hydrophobic cavities requiring higher LogP for binding.
Hydrogen bond donors (HBD) and acceptors (HBA): PPI inhibitors maintain similar H-bonding capacity to conventional drugs, but distributed over larger scaffolds.
Topological polar surface area (TPSA): Slightly lower than conventional drugs to maintain membrane permeability despite increased size.
Rotatable bonds (ROTB): Indicates flexibility needed to adapt to PPI binding surfaces, typically 5-8 bonds.
Aromatic rings (AROM): Count of aromatic systems. PPIs often involve π-π stacking, with optimal values around 2-3 rings.

Scoring algorithm

Each property gets converted to a desirability score using an asymmetric double sigmoid function:

$d(x) = a + \frac{b}{1 + \exp\left(-\frac{x - c + \frac{d}{2}}{e}\right)}\left[1 - \frac{1}{1 + \exp\left(-\frac{x - c - \frac{d}{2}}{f}\right)}\right]$

The sigmoid parameters ( $a, b, c, d, e, f$ ) were fitted to iPPI-DB compounds, creating peak desirability at property values common in successful PPI drugs.

The final QEPPI score combines these via weighted geometric mean:

$\text{QEPPI} = \exp\left(\frac{\sum_i w_i \ln(d_i)}{\sum_i w_i}\right)$

where weights $w_i$ were optimized to maximize discrimination between PPI modulators and non-modulators.

Understanding your results

Score interpretation

QEPPI returns a score between 0 and 1 for each molecule:

> 0.7: Excellent PPI drug-likeness. Properties align well with successful PPI inhibitors.
0.5-0.7: Good. Likely suitable for PPI targeting with minor optimization.
0.3-0.5: Moderate. May work for PPIs but consider property adjustments.
< 0.3: Poor PPI drug-likeness. Major redesign likely needed.

The default discrimination threshold is 0.52 (maximizes F-score), but we recommend using the continuous score for ranking rather than applying hard cutoffs.

Property-specific guidance

Review individual property values to understand why a molecule scored high or low:

High MW (>600 Da) but low QEPPI? Check if ALogP is too low. Large molecules need sufficient lipophilicity to maintain permeability.

Low score despite good MW/LogP? Examine TPSA and HBD/HBA. Excessive polarity prevents membrane permeability even with favorable size/lipophilicity.

Score near 0.5 boundary? These molecules are in borderline chemical space. Use QEPPI as one filter among many—combine with docking scores or experimental data.

Real-world example

Let's screen three known compounds:

ABT-263 (Navitoclax) - Bcl-2/Bcl-xL inhibitor in clinical trials:

1SMILES: CC1(C)CCC(C)(C)C2=CC(=C(C=C12)O)C3=C(C(=O)NC3=O)C4=CN=CC=C4NS(=O)(=O)C5=C(C=CC(=C5)NCC6=CN=C(C=C6)C7=CC=C(C=C7)Cl)S(=O)(=O)C(F)(F)F2Expected QEPPI: ~0.65 (good PPI drug-likeness)3MW: 974 Da (high but acceptable for PPI), ALogP: 7.2 (high lipophilicity for PPI interface)

Aspirin - Conventional COX inhibitor, not a PPI drug:

1SMILES: CC(=O)Oc1ccccc1C(=O)O2Expected QEPPI: ~0.15 (poor PPI drug-likeness)3MW: 180 Da (too small), ALogP: 1.2 (insufficient lipophilicity)

Venetoclax - FDA-approved Bcl-2 inhibitor for CLL:

1SMILES: CC1(C)CCC(C)(C)C2=C1C=C(C(=C2)C3=C(C(=O)NC3=O)C4=CC=CC=C4Cl)NS(=O)(=O)C5=CC6=C(C=C5)N(C=C6)CCN7CCOCC72Expected QEPPI: ~0.75 (excellent PPI drug-likeness)3MW: 868 Da, ALogP: 6.4, optimized for Bcl-2 hydrophobic groove

Comparison with QED and Rule-of-Four

Metric	QEPPI	QED	Rule-of-Four
Training data	1,007 PPI modulators	FDA oral drugs	39 PPI inhibitors
Scoring type	Continuous (0-1)	Continuous (0-1)	Binary (pass/fail)
MW optimum	500 Da	350 Da	>400 Da
LogP optimum	4.8	2.7	>4
AUC for PPIs	0.789	0.362	~0.65
Best for	PPI screening	General drugs	PPI binary filter

QEPPI outperforms QED for PPI compounds because it captures their distinct chemical space. QED scores PPI drugs low because they violate conventional drug rules—exactly the behavior we want to avoid when screening PPI-focused libraries.

Limitations

No structural alerts: QEPPI doesn't filter PAINS (pan-assay interference compounds) or reactive groups. Run separate filters for these.

Macrocycles underrepresented: Natural product PPI inhibitors and large macrocycles (>800 Da) may score lower than deserved due to limited training examples.

Target-agnostic: All PPI targets treated equally. Some interfaces (bromodomains) favor higher scores; others (transthyretin dimers) favor lower scores. Consider developing target-specific thresholds.

2D properties only: Doesn't account for stereochemistry or conformational flexibility, both important for PPI binding.

Not a guarantee: High QEPPI means favorable properties for PPI targeting, not guaranteed binding. Combine with docking or experimental validation.

Workflow integration

Step 1: Virtual screening Screen large compound libraries (ChEMBL, ZINC) with QEPPI to enrich for PPI-favorable molecules before expensive docking simulations.

Step 2: Lead optimization As you modify lead compounds, track QEPPI scores to ensure modifications maintain PPI drug-likeness. Aim to keep scores > 0.5.

Step 3: ADMET assessment Compounds passing QEPPI screening should proceed to ADMET-AI for toxicity and pharmacokinetics prediction.

Step 4: Experimental validation QEPPI identifies molecules worth testing—biochemical or cellular assays provide ground truth.

FAQ

Is QEPPI free to use?

QEPPI analysis costs 1 credit per job on ProteinIQ. Each job can analyze multiple molecules (paste SMILES line-by-line or upload a file), making batch screening cost-effective.

What input formats are supported?

QEPPI accepts:

SMILES strings: One per line in the text box
Tab-separated: name\tSMILES format for labeled compounds
File uploads: .smi, .csv, .txt, .sdf files
PubChem IDs: Fetch compounds directly from PubChem

What do I do with molecules that score 0.45-0.55?

These are in borderline chemical space. Use QEPPI as one data point among several—combine with:

Docking scores against your PPI target
Synthetic accessibility (can you make it?)
Similarity to known actives
Expert medicinal chemistry assessment

Don't use QEPPI alone for binary accept/reject decisions at borderline scores.

Can I use QEPPI for covalent PPI inhibitors?

QEPPI was trained on non-covalent PPI modulators. Covalent inhibitors may score differently because they rely on reactive electrophiles not captured in the model. Use QEPPI for general drug-likeness but don't rely on it for covalent warhead optimization.

How does QEPPI handle stereochemistry?

It doesn't. QEPPI calculates 2D molecular descriptors from SMILES, ignoring stereochemistry. If you have stereoisomers, they'll receive identical scores. This is a known limitation—use other tools for stereochemistry-dependent properties.

My known PPI inhibitor scored below 0.5. Why?

Several reasons:

Target specificity: Some PPI families favor lower scores (e.g., small interfaces like transthyretin dimers)
Macrocycles: Large natural products may be undervalued
Covalent binders: Not well-represented in training data
Outliers: QEPPI reflects average PPI chemical space; successful outliers exist

QEPPI is a statistical model, not absolute truth. Experimental data trumps computational predictions.

Can I screen millions of compounds?

Yes. QEPPI processes ~10 molecules per second, making large-scale screening feasible. Upload files with thousands of SMILES for batch processing. The lightweight algorithm (pure RDKit calculations) scales efficiently.

Should I filter out molecules with QEPPI < 0.3?

Depends on your screening funnel. If you have millions of compounds, filtering at 0.4-0.5 is reasonable to reduce downstream computational cost. If you have hundreds of compounds from focused libraries, review all scores—some successful PPI drugs fall below typical thresholds.

How do I cite QEPPI?

Kosugi, T., & Ohue, M. (2021). Quantitative estimate index for early-stage screening of compounds targeting protein-protein interactions. International Journal of Molecular Sciences, 22(20), 10925. https://doi.org/10.3390/ijms222010925

References

Bickerton, G. R., Paolini, G. V., Besnard, J., Muresan, S., & Hopkins, A. L. (2012). Quantifying the chemical beauty of drugs. Nature Chemistry, 4(2), 90-98. https://doi.org/10.1038/nchem.1243

Morelli, X., Bourgeas, R., & Roche, P. (2011). Chemical and structural lessons from recent successes in protein-protein interaction inhibition (2P2I). Current Opinion in Chemical Biology, 15(4), 475-481. https://doi.org/10.1016/j.cbpa.2011.05.024