Create a named analysis, choose training features, optionally restrict perturbations, optionally upload a matrix CSV, and launch the job.
Analysis name
Dataset
Loading datasets…
Available response sets
Features to use for training
Choose whether preprocessing should use the top variable features globally, or an explicit list of feature names. Explicit feature names may be delimited by newlines, commas, or semicolons, and server-side preprocessing will match against cleaned feature names.
Feature selection mode
Number of top variable features
Rank variable features using
These are the globally selected features used during preprocessing for all downstream models. Reference mode ranks features by variability in the reference training dataset. When a prediction cohort is provided, selected features are also restricted to features available for prediction.
Perturbations to train
Paste or upload a plain-text list of perturbation names. Delimiters can be newlines, commas, or semicolons. Server-side preprocess will match case-insensitively using the internally configured tag for the selected response set, for example CTNNB1 → ko_ctnnb1. It will not fuzzy-match near names such as ctnnb1 to ko_ctnnb13.
Train all models
Default mode. When enabled and no explicit perturbation list is uploaded, the backend will train all resolved perturbations for this response set.
No explicit perturbation list provided. The backend will train all resolved perturbations.
Random seed
Controls the deterministic cross-validation splits and XGBoost randomness for every perturbation. The backend derives a stable perturbation-specific seed, so selecting a different set or order of perturbations does not change a model.
Generate shuffled-response null models
When enabled, PowerUp also uses the random seed above to derive a separate stable response-permutation seed for each perturbation. Use different random seeds for repeated null jobs.
Paste perturbation list
Or upload perturbation list file
Current training-target mode
No explicit perturbation list provided. The backend will train all resolved perturbations for this response set.
Optional reference cell-line subsets
These lists apply only to the reference dataset. Training IDs restrict which reference cell lines are eligible for training. Test IDs are appended to the prediction cohort and excluded from training when applicable. Overlap is allowed; the backend resolves final training IDs as set difference.
Reference training cell lines
Optional. If provided, the backend will train only on these reference IDs, except any that also appear in the test list.
Reference test cell lines
Optional. If provided, matched reference rows will be appended to the prediction cohort and excluded from training when applicable.
Expression source
Matrix CSV file
Choose a CSV file
Optional. If provided, it is stored as matrix.csv and used for sample predictions. Leave it empty, with no reference test samples, to run a training-only job.
No file selected yet.
Models to train
using the selected dataset
Training features
top variable features ranked from the reference training dataset
Prediction cohort
No sample predictions will be generated
Want to run this analysis?
Public visitors can explore and configure this workflow, but new analyses require access through the HTS Lab.
