Create REINVENT4 Job

Configure molecular generation with QSAR optimization

Basic Information

Job Name *

Descriptive name for this generation job

QSAR Model Selection *

Note: Only publicly available QSAR models can be used for predictions. If you want to add your QSAR model to the publicly available list, please contact Dr Salah Alshehade.

Selected Model Details

Model Name:: -
Algorithm:: -
Created By:: -
Created Date:: -
Status:: -

Validation R²:: -
Training R²:: -
RMSE:: -
Description:: -

Tip: Higher R² values indicate better model performance. Choose a model trained on data similar to your target molecules.

Description

Optional description of this job's objectives

Generation Configuration

Generation Mode *

Reinvent: De novo design
Libinvent: Scaffold decoration
Linkinvent: Linker design
Mol2Mol: Optimization

Learning Algorithm *

Transfer Learning: Fast convergence
Reinforcement Learning: Best optimization
Curriculum Learning: Staged learning

Target Molecules *

Number of molecules to generate

Max Steps *

Maximum optimization steps

Batch Size *

Molecules per batch (typically 128)

Scoring Configuration

QSAR Weight * 0.60

Weight for QSAR prediction score (0.0 - 1.0)

QED Weight * 0.40

Weight for drug-likeness (QED) score (0.0 - 1.0)

Note: QSAR Weight + QED Weight must sum to approximately 1.0

Target Score

Optional target composite score to achieve

Diversity Threshold *

Minimum similarity threshold for diversity (0.0 - 1.0)

Cancel

Back to REINVENT4 Dashboard

How It Works

Generation

REINVENT4 uses AI to generate novel molecule structures (SMILES) based on your configuration.

QSAR Evaluation

Each generated molecule is evaluated by your selected QSAR model to predict biological activity.

Scoring

Scores are combined (QSAR + QED) to create a composite fitness score for each molecule.

Learning

REINVENT4's neural network learns from high-scoring molecules and generates improved candidates.

Optimization

Over multiple iterations, the system converges toward molecules with optimal properties.

Tip: Higher QSAR weight focuses on activity prediction, while higher QED weight emphasizes drug-likeness.