Proteomics Preprocessing Playbook: Missingness + Normalization

A Simple End-to-End Workflow

Use this order to keep results reproducible and biologically meaningful.

Step 1. Audit Data Quality

1. Inspect missingness per sample and per protein.
2. Remove clear outlier samples.
3. Flag proteins with very high missingness.

Step 2. Log Transform Intensities

Log2 transform stabilizes variance and improves comparability.

df_log2 <- log2(df + 1)

Step 3. Normalize Across Samples

Start with median normalization and evaluate. Move to quantile/VSN if needed.

Step 4. Impute Missing Values

Choose method based on likely mechanism:

1. MAR/MCAR: kNN or random forest.
2. MNAR (below detection): left-censored strategy.

Step 5. Run Sensitivity Checks

Repeat differential analysis with at least two imputation approaches and compare overlap in top proteins.

# Example: compare top hits across two methods
top_a <- head(results_method_a$protein[order(results_method_a$p_adj)], 50)
top_b <- head(results_method_b$protein[order(results_method_b$p_adj)], 50)
length(intersect(top_a, top_b))

Step 6. Report What You Did

Document:

1. Filtering threshold(s).
2. Normalization method.
3. Imputation method and parameters.
4. Sensitivity analysis outcome.

Key Takeaway

A transparent preprocessing pipeline is more valuable than chasing a single “perfect” method. Consistency, justification, and validation are what make findings trustworthy.