First challenge: Establishing purpose-built models using MCIA

The MICE (Multiple Imputation by Chained Equations) method was employed to replace missing data values in the harmonized dataset (Figure 2D). Specifically, transcriptome data was utilized to impute missing values in other data modalities through the application of MICE. We utlilized MICE imputed data to construct models. We implemented MCIA using the mbpca function in the mogsa package. We generated 10 low-dimension multi-omics factor scores for training datasets. Each multi-omics factor was derived through a linear combination of the original features (e.g., genes or proteins) extracted from the input data. Subsequently, global scores were computed for the test dataset, capturing the overall representation or summary of the data in relation to the underlying factors identified from the training data. This was accomplished by utilizing factor loadings from the training dataset and feature scores from the test dataset. For each task, we constructed a prediction model utilizing a general linear model with lasso regularization using the glmnet library. We used the feature scores as input data and the prediction task values as response variables, generating separate predictive models for each task.