Synthetic augmentation of cancer cell line multi-omic datasets using unsupervised deep learning

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Integrating diverse biological data is crucial for understanding cancer but faces challenges like data heterogeneity and sparsity.
• The Cancer Dependency Map (DepMap) provides valuable data but can be incomplete.

Purpose of the Study:

• To develop an unsupervised deep learning model for integrating and augmenting multi-omic data within the DepMap.
• To enhance the completeness and utility of the DepMap for cancer research.

Main Methods:

• Introduction of MOSA (Multi-Omic Synthetic Augmentation), an unsupervised deep learning model.
• Utilizing orthogonal multi-omic information to generate synthetic molecular and phenotypic profiles.
• Application of SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• Achieved a 32.7% increase in multi-omic profiles, creating a complete DepMap for 1523 cancer cell lines.
• Enhanced statistical power, enabling the discovery of novel drug resistance mechanisms.
• Improved identification of genetic associations and refined cancer cell line clustering.

Conclusions:

• MOSA effectively integrates and augments multi-omic data, significantly enhancing the DepMap.
• The model aids in uncovering complex biological mechanisms and identifying potential biomarkers for drug and gene dependencies.
• This approach is vital for developing strategies to prioritize cancer targets and advance precision oncology.