Integrative Prediction Strategy (IPS): Blood-based Prediction of Brain Gene Expression

main @ d0bae6b

Integrative Prediction Strategy (IPS) predicts brain gene expression from blood-derived features using pre-trained machine learning models. The workflow integrates multiple feature selection strategies (unsupervised and supervised) and evaluates model performance across cross-validation folds.

Workflow Steps:

1. Load matched blood and brain gene expression data.
2. Apply feature selection:
   • Unsupervised feature selection (features selected without using the target brain gene)
   • Supervised feature selection (features selected based on association with the target brain gene)
3. Load pre-trained models (5-fold cross-validation).
4. Select prediction mode: best, top10, or all models.
5. Generate predicted brain expression values.
6. Evaluate predictions (correlation with observed expression).
7. Save outputs to file.

Inputs: Blood features and matched brain expression data, pre-trained models, selected feature sets, cross-validation fold assignments.

Outputs: Predicted brain expression values, correlation metrics, model-specific prediction results.

Resource Identification: RRID: SCR_027608