Proteomizer: Leveraging the Transcriptome-Proteome Mismatch to Infer Novel Gene Regulatory Relations.

Area of Science:

• Multi-omics integration
• Computational biology
• Deep learning in genomics

Background:

• Transcriptomic (Tx) and proteomic (Px) profiles show modest correlation, limiting Tx data's utility for protein abundance.
• Existing methods struggle to accurately infer Px from Tx and miRNomic (Mx) data.

Purpose of the Study:

• Introduce Proteomizer, a deep learning platform to infer Px from Tx and Mx profiles.
• Evaluate Proteomizer's performance in improving differential gene expression analysis.
• Investigate Tx-Px discrepancies using explainable AI (XAI).

Main Methods:

• Trained Proteomizer on 8,613 matched Tx-Mx-Px samples from TCGA and CPTAC.
• Utilized Monte Carlo simulations to assess proteomization impact on differential expression.
• Applied XAI techniques and compared predictions against a biological knowledge graph.

Main Results:

• Achieved a record Tx-Px correlation of r=0.68.
• Significantly improved differential gene expression accuracy (up to 62-fold p-value precision increase).
• XAI identified regulatory relations with ROC-AUC of 0.74 for miRNA-gene downregulation predictions.

Conclusions:

• Proteomizer represents a state-of-the-art tool for multiomic integration.
• Performance is dataset-dependent, highlighting the need for careful validation.
• XAI provides interpretable insights into transcript-protein level discrepancies.