ToxPred 2.0 (Toxicity prediction)

Screen compounds for structural toxicity alerts using PAINS filters, BRENK filters, and NIH filters for safer drug discovery.

Job name

Molecule

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

Related tools

ADMET-AI

Predict ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) properties from SMILES strings using machine learning models trained on Therapeutics Data Commons datasets.

Admetica

Predict 22 ADMET properties from SMILES strings with the upstream Admetica Chemprop models from Datagrok.

Brenk filter

Identify toxic, reactive, and pharmacokinetically problematic molecular fragments using structural alert patterns

eToxPred

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.

Lead-likeness filter

Screen for lead-like compounds using stricter molecular descriptor criteria than Lipinski or Veber rules for early-stage drug discovery

PAINS filter

Screen compounds for Pan-Assay INterference patterns that cause false positives in biological assays

QEPPI

Quantitative estimate for protein-protein interaction inhibitor potential. Evaluates drug-likeness for compounds targeting PPIs.

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 5

Lipinski's Rule of Five predicts whether compounds will be orally bioavailable by evaluating molecular weight, LogP, hydrogen bond donors, and acceptors.

SPRINT

SPRINT (Structure-aware PRotein ligand INTeraction) predicts drug-target interactions using co-embedded protein and ligand representations. Screen thousands of compounds against a protein target in seconds.

What is toxicity prediction?

Toxicity prediction screens compounds for problematic molecular fragments before expensive biological testing. Rather than predicting specific toxicity endpoints, structural alert filters match known toxic, reactive, or interference-prone substructures curated from decades of medicinal chemistry. This makes them useful throughout drug discovery, from filtering compound libraries and triaging virtual screening hits to tracking safety profiles during lead optimization.

Keep in mind that structural alerts are context-independent: a flagged substructure may be benign in certain scaffolds, and many approved drugs trigger alerts (aspirin flags Brenk patterns for its acetyl group). They also cannot cover every toxic structure or explain the specific toxicity mechanism. Alerts work best as one layer in a broader safety assessment alongside druglikeness filters like Lipinski's Rule of Five, Veber's Rule, and the Lead-Likeness Filter, not as a standalone pass/fail gate.

How ToxPred 2.0 works

ToxPred 2.0 is ProteinIQ's own scoring algorithm that combines three established structural alert filter sets into a single weighted risk score:

PAINS filters (weight: 0.2) flag assay interference compounds
Brenk filters (weight: 0.3) flag toxic, reactive, and pharmacokinetically problematic fragments
NIH filters (weight: 0.25) flag problematic functional groups from NIH/MLPCN screening campaigns

Each filter set runs its SMARTS patterns against the input molecule. Alerts from all enabled filters combine into a risk score from 0 to 1:

\text{Risk Score} = \min\!\Big(1.0,\;\sum_{i} w_i \cdot n_i\Big)

where $w_i$ is the per-alert weight for filter $i$ and $n_i$ is the number of detected alerts. Each alert adds its weight to the total, and the score is capped at 1.0. For example, a compound with 2 PAINS alerts and 1 Brenk alert scores $2 \times 0.2 + 1 \times 0.3 = 0.7$ .

Compounds are then classified based on their score:

Low risk (0.0 to 0.3): Few alerts, suitable for development
Moderate risk (0.3 to 0.7): Some alerts present, enhanced monitoring recommended
High risk (0.7 to 1.0): Multiple alerts, structural changes or deprioritization recommended

A compound is classified as Safe only when it triggers zero alerts across all filter sets.

How to predict toxicity online

ProteinIQ screens compounds for structural toxicity alerts using PAINS, Brenk, and NIH filters directly in the browser with no installation required.

Enter SMILES strings in the text area, one per line. Compound names can be included using tab-separated format (aspirin\tCC(=O)Oc1ccccc1C(=O)O). File upload is supported for .smi, .csv, and .smiles formats, and compounds can also be fetched from PubChem by name or CID.

Settings

All three filter sets run by default. To focus on a specific category, toggle off the ones not needed.

Setting	Description
`PAINS filter`	Screen for pan-assay interference compounds (PAINS A, B, C). Default: on.
`Brenk filter`	Screen for toxic, reactive, and pharmacokinetically problematic fragments. Default: on.
`NIH filter`	Screen for compounds with problematic functional groups (NIH/MLPCN). Default: on.

If all three are toggled off, the tool falls back to running all of them.

Results

Each compound receives a per-filter alert count, a combined risk score, and a categorical toxicity classification.

Column	Description
`Name`	Compound name (if provided) or SMILES
`SMILES`	Input SMILES string
`Toxicity`	Categorical classification: `Low`, `Moderate`, or `High`
`Risk score`	Weighted composite score from 0 (no alerts) to 1 (maximum risk)
`Total alerts`	Sum of alerts across all enabled filters
`PAINS alerts`	Number of PAINS substructure matches
`Brenk alerts`	Number of Brenk substructure matches
`NIH alerts`	Number of NIH substructure matches
`PAINS patterns`	Names of matched PAINS patterns
`Brenk patterns`	Names of matched Brenk patterns
`NIH patterns`	Names of matched NIH patterns