Input
1 credit
Output
Configure input settings on the left, then click "Submit job"orLoad an example (it's free)
PPI inhibitor screening
PPI inhibitor screening
Predict ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) properties from SMILES strings using machine learning models trained on Therapeutics Data Commons datasets.
Predict 22 ADMET properties from SMILES strings with the native Admetica Chemprop models from Datagrok.
Identify toxic, reactive, and pharmacokinetically problematic molecular fragments using structural alert patterns
Predict toxicity and synthetic accessibility of small molecules using machine learning. eToxPred combines toxicity risk assessment with synthetic accessibility scoring to help prioritize drug candidates.
Screen for lead-like compounds using stricter molecular descriptor criteria than Lipinski or Veber rules for early-stage drug discovery
Screen compounds for Pan-Assay Interference patterns that cause false positives in biological assays
Screen compounds for structural toxicity alerts using PAINS, Brenk, and NIH filters. For focused screening, see PAINS Filter, Brenk Filter, or Veber's Rule.
Veber's Rule predicts oral bioavailability by evaluating molecular weight, LogP, hydrogen bond donors/acceptors, and rotatable bonds
Lipinski's Rule of Five predicts whether compounds will be orally bioavailable by evaluating molecular weight, LogP, hydrogen bond donors, and acceptors.
AF2BIND predicts ligand-binding residues from a protein structure using AlphaFold2 pair representations and a 20-residue bait sequence.
QEPPI (Quantitative Estimate Index for Compounds Targeting Protein-Protein Interactions) is a drug-likeness scoring function for molecules that target protein-protein interactions. It returns a score from 0 to 1 calibrated against 1,007 experimentally validated PPI modulators from the iPPI-DB database.
Standard drug-likeness rules like Lipinski's Rule of 5 were designed for enzyme inhibitors and reject most successful PPI drugs. PPI binding surfaces are large and flat, requiring bigger and more lipophilic molecules to disrupt them. QEPPI captures this distinct chemical space: its AUC for separating PPI inhibitors from decoys is 0.789, compared to 0.362 for QED on the same set.
ProteinIQ runs QEPPI calculations in the cloud on any number of compounds at once. Paste SMILES strings into the input box (one per line, optionally tab-separated with names), upload a file, or fetch compounds directly from PubChem by name or CID. Each molecule receives a QEPPI score and the seven underlying molecular descriptors that determine it.
| Input | Accepted formats |
|---|---|
| SMILES text | One SMILES per line; optionally name\tSMILES for labeled output |
| File upload | .smi, .smiles, .sdf, .csv, .txt |
| PubChem fetch | Compound name or CID |
Each row in the output corresponds to one input molecule.
| Column | Description |
|---|---|
QEPPI score | Overall PPI drug-likeness score (0 to 1). Higher is more favorable. |
MW | Exact molecular weight in Da |
ALogP | Crippen partition coefficient (estimated LogP) |
HBD | Hydrogen bond donors |
HBA | Hydrogen bond acceptors |
TPSA | Topological polar surface area (Ų) |
Rotatable bonds | Count of rotatable bonds |
Aromatic rings | Count of aromatic ring systems |
The default discrimination threshold from the 2021 publication is 0.52, chosen to maximise F-score on the iPPI-DB test set. The continuous score is more useful than a hard cutoff for ranking and prioritization:
| Score | Interpretation |
|---|---|
| > 0.7 | Strong PPI drug-likeness, properties align with successful PPI inhibitors |
| 0.5–0.7 | Good, minor property adjustments may help |
| 0.3–0.5 | Moderate, consider structural modifications |
| < 0.3 | Poor PPI drug-likeness, major redesign likely needed |
The descriptor columns explain why a molecule scored as it did. High MW with low ALogP often drives a low score: PPI inhibitors need both to maintain permeability across a large scaffold. Low TPSA with acceptable HBD/HBA is typical of good PPI compounds. An aromatic ring count of 2–3 tends to be optimal for π-stacking with PPI interfaces.
Molecules scoring near 0.5 sit in borderline chemical space. Combine the score with docking results, synthetic accessibility, and experimental data rather than treating QEPPI alone as a binary accept/reject filter.
| 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 |
Use QEPPI when targeting protein-protein interactions specifically: Bcl-2 family proteins, MDM2/p53, bromodomains, or similar interfaces. Lipinski's Rule of 5 is appropriate for conventional enzyme and receptor targets where oral bioavailability under classical criteria is the priority. ADMET-AI addresses absorption, distribution, metabolism, excretion, and toxicity rather than drug-likeness; it fits later-stage lead optimization after QEPPI has filtered a virtual library. Molecular Descriptors provides the raw physicochemical values without a composite score, useful for building custom QSAR models.
QEPPI calculates seven molecular descriptors from SMILES, converts each to a desirability score using an asymmetric double sigmoid function (ADS), then combines them into a final score via weighted geometric mean.
The ADS function fitted to iPPI-DB compounds creates a peak desirability at property values common in successful PPI drugs:
The six coefficients per descriptor (–) were fitted to the iPPI-DB dataset. The final score is a weighted geometric mean over all seven desirability values:
Weights were optimised to maximise discrimination between PPI modulators and non-modulators. The default weight set is the median-optimised configuration from the publication.
The calculation is pure 2D: all descriptors come from SMILES without 3D conformer generation. Stereoisomers receive identical scores. QEPPI also has no structural alert component, so PAINS compounds and reactive groups are not penalised. Run PAINS Filter or Brenk Filter in parallel for those screens.
Macrocycles and large natural product PPI inhibitors (above ~800 Da) tend to score lower than their activity warrants, because they are underrepresented in the iPPI-DB training data. Covalent PPI inhibitors are similarly out-of-distribution; the model has no knowledge of reactive warhead chemistry.
| Compound | SMILES prefix | Expected score | Why |
|---|---|---|---|
| Aspirin | CC(=O)Oc1ccccc1... | ~0.15 | MW 180 Da, ALogP 1.2: too small and polar for PPI |
| Navitoclax (ABT-263) | CC1(C)CCC(C)(C)... | ~0.65 | MW 974 Da, ALogP 7.2: good PPI profile despite large size |
| Venetoclax | CC1(C)CCC(C)(C)... | ~0.75 | MW 868 Da, ALogP 6.4: FDA-approved Bcl-2 inhibitor |
Aspirin scores poorly not because it is a bad drug, but because it occupies chemical space designed for traditional enzyme inhibition. Venetoclax and Navitoclax score well because they were deliberately optimised for the large, hydrophobic Bcl-2 binding groove.
QEPPI is most useful at the front of the screening funnel, before expensive computations: