SPACE

SPACE generates network-based cross-species protein embeddings from STRING protein interaction data to support protein function and subcellular localization prediction.

Key Features:

• Network-Based Embeddings: Utilizes the STRING database containing protein interaction data for 1322 eukaryotes to produce embeddings that capture network relationships beyond sequence information.
• Cross-Species Alignment: Creates species-specific network embeddings and aligns them using orthology relations to enable direct cross-species comparisons.
• Complementarity to Sequence Embeddings: Network embeddings are complementary to sequence-based embeddings, with alignment informed by sequence-based orthology.
• Validation and Performance: Validated on subcellular localization and protein function prediction by training logistic regression classifiers on aligned network and sequence embeddings, achieving performance comparable to state-of-the-art deep-learning methods.

Scientific Applications:

• Protein Function Prediction: Supports cross-species prediction of protein function by integrating network-derived and sequence-derived embeddings.
• Subcellular Localization Prediction: Improves cross-species prediction of protein subcellular localization using aligned network and sequence embeddings.

Methodology:

Integrates protein interaction networks from STRING with orthology information, generates species-specific network embeddings and aligns them using orthology relations, and trains logistic regression classifiers on aligned network and sequence embeddings for validation.