Phase 4: Predictive modelling

In this phase, you will create models to predict the responder status at 6 months after infection from early timepoint data (day 28)

Upload the processed dataset from Phase 4, then configure and run predictive models in PANDORA

1) Setup prediction task

Step 1. Upload the processed dataset

1. Navigate to Workspace

2. Upload the covid_pitch_day28_predictors_month6_outcome.csv file onto Workspace

3. Select the uploaded dataset

---

Step 2. Navigate to Predictive -> Start

---

Step 3. Configure analysis properties:

1. Use the toggle switch to select all columns as the Predictor variables

2. Use the Exclude predictors option to exclude Donor ID

3. Select Responder column for Response

   1. You will not see the Responder variable when you scroll since it is beyond the first 50 variables, so type out the variable name and it will appear

4. Set the Training/Testing dataset partition to 75% training (and hence 25% testing)

5. Select Preprocessing options center, scale, and medianImpute

---

Step 4. Select packages for your predictive models.

1. For this example, we will select families of algorithms especially suitable for biomedical data:

   1. L1 Regularization : Also known as LASSO

   2. L2 Regularization : Also known as ridge penalty

   3. Sparse Partial Least Squares: This method was used in the paper

   4. Random Forest

   5. Support Vector Machines

For more details about these algorithm families, visit the subpage Algorithms for biomedical data

Caution: Running too many models simultaneously on a personal computer may significantly increase processing time, and computationally intensive models may fail due to Timeout limit

Experimental options

When creating your own predictive models, you can experiment with:

• Packages: PANDORA has 200+ packages for predictive models. You can choose other families of algorithms or select individual algorithms.

• Training/Testing dataset partition: Different models perform better in different partitions, and experimenting with this parameter can help generate the best model.

• Multi-set Intersection

• Feature Selection

2) Run analysis

Step 1. Click the Validate data button at the bottom right of the screen

---

Step 2. Click the Process button on the pop-up

Your predictive modelling has started! You can monitor progress on the PANDORA Dashboard

You have successfully configured the predictive models using PANDORA. Once your models have completed processing, you're ready to interpret the results and evaluate model performance in the next phase.