Guide
A practical walkthrough of the PreditX® workflow.
From target selection to ranked screening outputs.
PreditX® guides users through a target-based AI drug discovery workflow: train models from public bioactivity data, review model outputs, screen molecules, and prioritize candidates using consensus prediction and ADMET-supported triage.
The complete workflow
The application is organized into four user-facing steps. Each step has a dedicated page so non-technical users can understand where they are in the process.
Train models
Enter a protein target, select descriptors, and choose the machine learning models you want PreditX® to train.
Review training results
Inspect training status, model outputs, performance files, and determine whether the trained models are ready for screening.
Start screening
Select a completed training run and screen pasted molecules, uploaded SMILES, or the PreditX® internal compound database.
Review screening results
Open ranked screening outputs, consensus predictions, CSV files, reports, and ADMET workbooks.
Step 1 — Train models
Model training begins with a biological target. PreditX® retrieves and prepares target-specific bioactivity data, calculates molecular features, and trains selected models.
Target input
Start by entering a protein target. The platform uses public target and bioactivity resources to assemble a target-specific training dataset.
- Protein target selection
- Automated data retrieval
- Target-specific model generation
Descriptor selection
Choose how molecules are represented for machine learning.
- Morgan fingerprints
- Mordred descriptors
- Combined fingerprints and descriptors
Model selection
Select one or more supervised machine learning models for training.
- Random Forest, SVM, XGBoost
- Logistic Regression, k-NN, MLP
- Tree-based, linear, and baseline models
Step 2 — Training results
After training, review the generated outputs before moving into screening.
What to check before screening
A completed training run means the model-training workflow has finished and can be used for downstream molecule screening. Review the generated outputs, trained models, and any available performance information before launching screening.
- Training status and progress
- Models trained for the selected target
- Generated output files and reports
- Readiness for downstream screening
Step 3 — Start screening
Screening starts from a completed training run. PreditX® can evaluate molecules from several sources.
Paste molecules
Paste SMILES directly into the interface for quick screening of smaller lists.
Upload a molecule file
Upload a SMILES file for larger screening batches.
Use the internal database
Screen against the PreditX® internal database of purchasable, PAINS-filtered compounds to generate a ranked candidate shortlist.
Step 4 — Screening results & ADMET
The final output is designed to support decision-making, not replace experimental validation.
Consensus ranking
When multiple models are available, PreditX® can combine model outputs to support more robust prioritization.
Ranked candidates
Screening results are returned as ranked molecules with prediction scores, classes, and model-support information.
ADMET triage
ADMET outputs help users review drug-likeness, physicochemical properties, structural alerts, risk bands, and recommended next actions.
Important interpretation note
PreditX® is a decision-support platform for early-stage compound prioritization. Predicted activity scores should be interpreted as prioritization signals, not as experimental potency values. Final decisions should include medicinal chemistry review, confirmatory assays, and, when relevant, structure-based assessment.