IsoResolve

IsoResolve applies domain adaptation to predict isoform-specific functions by aligning gene-level and isoform-level features in a shared latent variable (LV) space to transfer gene functional annotations to isoforms arising from alternative splicing.

Key Features:

• Isoform-level function prediction: Focuses on predicting functions of isoforms produced by alternative splicing rather than treating genes as uniform entities.
• Domain adaptation (DA) integration: Integrates domain adaptation techniques with existing gene function prediction models to enable cross-domain information transfer.
• Source/target domain conceptualization: Treats gene-level features as a source domain and isoform-level features as a target domain for transfer learning.
• Latent variable (LV) projection: Projects gene and isoform domain features into a shared LV space to facilitate alignment of their distributions.
• Distribution alignment: Aligns LV distributions between domains to enable effective transfer of functional information from genes to isoforms.
• Performance benchmarking: Demonstrated improved prediction performance compared to five state-of-the-art methods.
• Case-study validation: Validated predictions with case studies involving genes that have known isoform-level functional annotations.

Scientific Applications:

• Functional genomics: Provides isoform-resolved functional annotations to refine gene function maps in functional genomics studies.
• Isoform annotation transfer: Transfers gene-level functional information to isoforms to improve annotation granularity for alternatively spliced products.
• Method benchmarking and evaluation: Serves as a comparative framework for evaluating isoform function prediction against state-of-the-art methods.
• Validation of isoform annotations: Supports case-study validation of isoform-level functional assignments using genes with known isoform annotations.

Methodology:

IsoResolve projects gene- and isoform-level features into a shared latent variable (LV) space, uses domain adaptation to align LV distributions between the gene (source) and isoform (target) domains, and integrates these DA-aligned representations with existing gene function prediction models to transfer functional information to isoforms.