Multi-Plex Tox Screens

​

The Problem You Are Trying to Solve

“I have a set of lead compounds, and I want to determine whether they are likely to be toxic by screening them against a panel of protein targets known to contribute to toxicity in my indication.” At the lead optimization stage, toxicity is one of the most common and costly causes of failure. Experimental tox panels are expensive and slow, and many liabilities only emerge late unless proactively assessed. Typical challenges include:

• Off-target binding that is not obvious from primary efficacy screens
• Toxicity arising from unintended protein interactions (e.g., ion channels, nuclear receptors, metabolic enzymes)
• Difficulty distinguishing true toxicity risk from noise in early data
• Lack of mechanistic insight into why a compound is toxic

This workflow is designed to computationally triage toxicity risk early, providing both predictions and mechanistic explanations to guide safer optimization.

​

Solution

This workflow combines target-based binding screens with AI-driven toxicity prediction to evaluate whether lead compounds are likely to engage known toxicity-associated proteins. The primary toxicity assessment chain is: Toxicity Target Panel Definition → Virtual Screening (Docking) → Interaction Analysis → ADMET-AI Toxicity Prediction → Virtual Cell Line models for Enterprise This layered approach allows users to:

• Screen leads against known toxicity-relevant proteins
• Identify off-target binding risks before experiments
• Understand which interactions may drive toxicity
• Prioritize compounds for safer optimization paths

Other engines (MD, FEP, quantum chemistry, retrosynthesis) can be integrated when deeper validation or redesign is required.

​

What Data Do I Need to Provide?

Required

• Lead compound structures (SMILES or CSV)
• A curated panel of toxicity-associated protein targets (PDBs or predicted structures)

Recommended

• Known toxicity benchmarks (e.g., hERG binders, CYP inhibitors)
• Experimental safety flags (if available)

Optional

• Multiple conformations or states of toxicity targets
• Membrane context (for ion channels or transporters)

​

Workflow

1. Define a Toxicity Target Panel

Begin by explicitly defining what toxicity means for your indication. Typical toxicity panels may include:

• Ion channels (cardiac channels, neuronal channels)
• Metabolic enzymes (CYP450s)
• Nuclear receptors (PXR, CAR, AhR)
• Transporters (P-gp, BCRP)
• Stress-response or apoptosis-related proteins

This step ensures the screen is hypothesis-driven, not generic. This step results in a curated, indication-relevant toxicity target panel.

2. Prepare Toxicity Targets for Screening

Each target must be docking-ready and biologically relevant. On Revilico, users typically:

• Upload or retrieve protein structures for each toxicity target
• Clean and standardize structures (remove irrelevant ligands, assign protonation)
• Define binding regions (known ligand sites or functional pockets)

For flexible or cryptic sites, Protein Water MD can later be used to generate alternate conformations. This gives a set of docking-ready toxicity targets.

3. Screen Leads Against Toxicity Targets

Now test whether leads bind where they shouldn’t. This step focuses on risk identification before optimization procedures using generative chemistry. Use Static Docking to:

• Rapidly screen each lead against the toxicity panel
• Identify strong or recurrent off-target binding signals
• Capture candidate binding poses for interpretation

What you’re looking for

• Unexpected strong predicted affinities
• Consistent binding across multiple toxicity targets
• Binding in functionally critical regions (e.g., pore, catalytic site)

This screen will give a target-by-compound off-target binding matrix that can be utilized to create a toxicity profile for your lead sets.

4. Analyze Interaction Patterns and Mechanisms

Not all binding equals toxicity, interpretation matters. Use Pharmacophore Analysis to:

• Identify interaction motifs driving off-target binding
• Compare these motifs to known toxicophores
• Distinguish promiscuous binders from target-specific interactions

This step answers:

• Is toxicity risk structural or incidental?
• Which parts of the molecule are responsible?

What results is a mechanistic hypothesis linking structure to toxicity risk.

5. Validate High-Risk Interactions with Dynamics (Optional)

For leads showing concerning signals, use Protein-Ligand MD to:

• Test whether off-target binding is stable over time
• Identify transient vs persistent interactions
• Eliminate docking false positives using extended dynamic time scales and different solvent conditions

This step is especially valuable for:

• Ion channels
• Flexible receptors
• Borderline docking results

This step will output a dynamic confirmation (or dismissal) of off-target binding risk, allowing for a more robust mechanistic understanding of off-target engagements.
This step is especially important for kinase targets which share homology with several other kinases biologically.

6. Predict System-Level Toxicity Endpoints

Complement target-based screening with data-driven prediction. Use Revilico’s ADMET-AI Engine to:

• Predict toxicity endpoints (hERG, DILI, AMES, LD50, etc.)
• Assess metabolic inhibition and clearance risks
• Surface confidence-scored alerts and red flags
• These models are trained on a wide expanse of experimental data to help get a picture of toxicity across different parameters

This step provides a global safety perspective beyond individual targets, and a consolidated toxicity risk profile per lead.

​

Results

• Early identification of potential toxicity liabilities
• Mechanistic insight into off-target interactions
• Reduced risk of late-stage failure
• Clear guidance on which leads to advance, redesign, or deprioritize
• The mechanistic insight gained from these engines will allow for better optimization of downstream compound sets especially when it comes to toxicity

​

Now What? I’ve identified toxicity risks, but what’s next?

Typical next steps include:

• Structural redesign to remove toxicophores
• Scaffold hopping or motif preservation workflows
• Focused experimental safety assays
• Re-screening optimized compounds against the toxicity panel

​

Why Revilico?

Revilico enables proactive toxicity assessment by integrating:

• Target-based virtual screening for off-target risk
• Interaction-level analysis for mechanistic clarity
• AI-driven toxicity prediction for system-wide safety insight
• Seamless iteration into redesign and optimization workflows

This allows teams to fail fast, learn early, and design safer molecules before toxicity becomes an expensive problem.