wenda_gpu: fast domain adaptation for genomic data.

Area of Science:

• Computational biology
• Genomics
• Machine learning

Background:

• Domain adaptation is crucial for developing predictive models with limited genomic data.
• Weighted elastic net domain adaptation effectively uses genomic features for data transferability.
• Existing methods are computationally intensive for large-scale genomic datasets.

Purpose of the Study:

• To develop a computationally efficient domain adaptation method for genomic data.
• To enable accurate prediction of cancer mutation status using limited sample sizes.
• To improve the transferability of predictive models in genomics.

Main Methods:

• Developed wenda_gpu, a GPU-accelerated implementation of weighted elastic net domain adaptation.
• Utilized GPyTorch for efficient model training on genomic data.
• Evaluated wenda_gpu's performance against the original wenda implementation and regular elastic net.

Main Results:

• wenda_gpu significantly reduces training time for genomic datasets, completing analyses within hours on a single GPU.
• The wenda_gpu method achieves comparable predictive performance to the original wenda implementation.
• wenda_gpu demonstrates superior performance in predicting cancer mutation status on small sample sizes compared to regular elastic net.

Conclusions:

• wenda_gpu offers a computationally feasible solution for domain adaptation in genomics.
• The tool facilitates improved predictive modeling for cancer mutation status with limited data.
• wenda_gpu enhances the application of domain adaptation techniques to genome-sized datasets.