ADMETlab 3.0 inputs & parameters

You provide compounds as SMILES — entered directly or imported from a previous job. The one setting that most affects what you get back is whether you enable full ADMET calculation: leave it off for fast structure processing, or turn it on to run the broad multi-endpoint prediction, keeping compounds in drug-like chemical space for reliable results.

Webapp input methods

The .bloom* webapp provides two input methods for ADMETLab3 molecular analysis:

Compound source selection

Manual Entry (default)

• Enter SMILES strings directly in the multiline text area
• One SMILES per line
• Maximum 100 molecules per job
• Molecule count displayed below input field

From Previous Job

• Import compounds from completed workflows
• Select from AiZynthFinder, REINVENT, ZINC, or PubChem outputs
• Automatic SMILES extraction from job results

SMILES input format

Enter SMILES strings in the text area, one per line:

CCO
CC(C)OC(=O)CC(=O)CSc1nc2c(cc1C#N)CCC2
C1=CC=C(C=C1)O

SMILES requirements

• Use canonical SMILES notation for consistent results
• Ensure valid valence states and bond orders
• Molecules should be within drug-like chemical space (typically MW < 1000 Da)
• Salt stripping is handled automatically when "Molecule Washing" is enabled

Webapp parameters

Analysis options

Toggle these options in the main interface:

Advanced options (expandable panel)

ADMET calculation options (when enabled)

• Use Feature-Based Models: Enable descriptor-based prediction models
• Include Uncertainty Models: Add confidence estimates to predictions

Image generation settings (when enabled)

• Width: Image width slider (100-1000 px, default: 300)
• Height: Image height slider (100-1000 px, default: 300)

Output configuration

• Output Filename Stem: Base name for output files (default: admetlab3_results)

Processing times

Estimates displayed in the webapp:

Validation and standardization

• Canonicalization, neutralization, and salt removal are on by default (via Molecule Washing toggle)
• Invalid structures are flagged with specific error reasons; valid rows continue processing
• Duplicate structures collapse to a single canonical SMILES with original_index retained

Configurable parameters

ADMETlab 3.0 predicts across 119 ADMET endpoints:

• Core endpoints: 88 established ADMET properties
• Extended endpoints: 31 new v3.0 endpoints including nephrotoxicity, neurotoxicity, ototoxicity, hematotoxicity, and others

The model is trained on 400,000+ entries and uses the DMPNN (Directed Message Passing Neural Network) architecture.

Tips

• Include an ID column to simplify joins downstream
• Enable "Calculate ADMET Properties" for full prediction; disable for faster structure processing
• Use image generation for visual compound libraries and reports
• Keep batches under 100 molecules for optimal performance

Read next