Step 1 — Model training

Create target-specific models for molecule screening.

Enter a biological target, choose how molecules should be represented, select the machine learning models to train, and submit an asynchronous job. Once training is completed, the trained models can be used to screen uploaded molecules, pasted SMILES, or the PreditX® internal database.

Training job flow

This page only starts model training. Screening comes after a model set is ready.

Define target and representation.

Train selected ML models asynchronously.

Use completed models for molecule screening.

Training setup

Complete the four steps below. The backend workflow remains asynchronous, so you can monitor progress later from the dashboard.

Authenticated

Biological target

Select or enter the target that defines your discovery question.

Target

Loading protein options...

Molecular representation

Choose how molecules should be converted into model-ready input features.

Mordred descriptors

Physicochemical and structural molecular descriptors.

Fingerprints

Structural fingerprints useful for substructure and similarity patterns.

Both

Combines descriptors and fingerprints for richer representation.

Machine learning models

Select the algorithms you want to train for this target.

Random Forest

SVM

Logistic Regression

XGBoost

Gradient Boosting

Extra Trees

Submit asynchronous job

Confirm your email and start model training.

Notification email

You can monitor progress from the dashboard. Email is used for notification and traceability.

Ready to start training?

After submission, the job will run in the background. You do not need to keep this page open.

What happens after submission?

PreditX® starts an asynchronous target-based training workflow. When the job is complete, you can use the trained models to screen molecules.

Training logic: Target → Evidence → Molecular representation → ML training → Model outputs → Screening-ready job

Target evidence

The selected target defines the bioactivity evidence used for model development.

Feature generation

Molecules are represented using fingerprints, Mordred descriptors, or both.

Model training

Selected algorithms are trained and evaluated for the target-specific prediction task.

Dashboard monitoring

Use the dashboard to check whether the asynchronous training job is queued, running, completed, or failed.

Screening step

After completion, use the trained models to screen uploaded molecules, pasted SMILES, or the internal database.

Train models