Across-Site Predictability of Metabolite Profiles

Data

• load_packages_names.R
• Raw metabolite data
• Covariates data: S-I_A covariates, S-I_B covariates, S-II covariates, S-III covariates.

Codes

• Data preprocessing:
  • Goal: Imputation and normalization to prepare data for statistical analysis. 
  • meta_preprocess.R
  • covar_preprocess.R

• R2 calculation: 
  • Goal: Calculate marginal R2 and conditional R2 for three sets of variables--demographics (D), metabolites (M), cohorts (C).
  • Results contains: Total R2, R2(D), R2(M), R2(C), R2(D|M,C), R2(M|D,C), R2(C|D,M) for all metabolites
  • Codes: rsq_subset.R

• Two-cohort study
  • Goal: Construct predictive model using training set (from S-I_B and S-II) and test the model using test set from from S-I_B and S-II plus all samples from S-I_A and S-III.
  • Step 1. Obtain training and test set from S-I_B and S-II: 2cohort_split_train_test.R
  • Step 2. Training model using training set: 2cohort_train_model.R
  • Step 3. Test model in Step 2 using test set:
    • Test using test set from S-I_B and S-II: 2cohort_test_testset.R
    • Test using samples in S-III: 2cohort_test_S-III.R
    • Test using samples in S-I_A: 2cohort_test_S-IA.R
  • Step 4. Measure variation using Averaged Squared Error (ASD)
    • Calculate deviation distance: 2cohort_calc_D.R
    • Calculate ASD: 2cohort_calc_ASD.R

• Three-cohort study
  
  • Goal: Construct predictive model using training set (from S-I_B, S-II and S-III) and test the model using test set from from S-I_B, S-II and S-III plus all samples from S-I_A.
  • Step 1. Obtain training and test set from S-I_B, S-II and S-III: 3cohort_split_train_test.R

• • Step 2. Training model using training set: 3cohort_train_model.R
  • Step 3. Test model in Step 2 using test set:
    • Test using test set from S-I_B, S-II and S-III: 3cohort_test_testset.R
    • Test using samples in S-I_A: 3cohort_test_S-IA.R
  • Step 4. Measure variation using Averaged Squared Error (ASD)
    • Calculate deviation distance: 3cohort_calc_D.R
    • Calculate ASD: 3cohort_calc_ASD.R

• One-cohort study 
  • Goal: Construct predictive model using training set (from S-I_B) and test the model using test set from from S-I_B plus all samples from S-I_A, S-II and S-III.

• • Step 1. Obtain training and test set from S-I_B: 1cohort_split_train_test.R

• • Step 2. Training model using training set: 1cohort_train_model.R
  • Step 3. Test model in Step 2 using test set:
    • Test using test set from S-I_B: 1cohort_test_testset.R
    • Test using samples in S-I_A, S-II and S-III: 1cohort_test_S-IA.R, 1cohort_test_S-II.R, 1cohort_test_S-III.R
  • Step 4. Measure variation using Averaged Squared Error (ASD)
    • Calculate deviation distance: 1cohort_calc_D.R
    • Calculate ASD: 1cohort_calc_ASD.R

• Testing with an additional cohort S-IV ( newly collected 253 samples)
  • Data: raw metabolite data meta_raw_bioivt_penn.txt
  • Codes:
    • Preprocess data from S-IV: preprocess_meta_S4.R 
    • Test using 2-cohort model: 2cohort_test_S-IV.R
    • Test using 3-cohort model: 3cohort_test_S-IV.R

• Figures
  • Histograms of observed, predicted and residuals from two-cohort model: Download HERE.

Publications