DECODE is introduced as a versatile deep learning framework designed to estimate cell-type and cell-state abundances across diverse biological datasets. Unlike previous algorithms limited to specific data types, this tool functions as a universal deconvolution solution for transcriptomic, proteomic, and metabolomic profiles. The system employs adversarial training to eliminate batch effects and contrastive learning to ensure high accuracy even when the reference data is incomplete or noisy. Extensive testing shows that it consistently outperforms existing methods across different diseases, species, and measurement platforms. Ultimately, the authors demonstrate its clinical utility by using it to reveal cellular composition changes in breast cancer metastasis and liver disease cohorts.

References:

• Zhao T, Liu R, Sun Y, et al. DECODE: deep learning-based common deconvolution framework for various omics data[J]. Nature Methods, 2026: 1-13.