Urgent.News

What's breaking now, across thousands of outlets.

AI

Lorentz Tangent Diffusion Adapter for Multi-label Prediction of Protein-Protein Interaction Types

Protein-protein interactions shape cellular organization and disease processes, but their prediction remains difficult because known interaction networks are incomplete, noisy and highly uneven across proteins. Hyperbolic graph learning is well suited to the hierarchical structure of such networks, yet most methods emphasize learning hyperbolic embeddings and pay less attention to how these…

Protein-protein interactions play a vital role in cellular organization and disease processes, yet predicting these interactions remains challenging due to incomplete, noisy, and uneven networks across proteins. Hyperbolic graph learning is adept at handling such hierarchical networks; however, most existing methods focus on learning hyperbolic embeddings while neglecting the propagation and decoding of these embeddings for predicting interaction types.

The Lorentz Tangent Diffusion Adapter (LTDA) addresses this gap by introducing a post-embedding module for predicting the type of interaction between a pair of proteins.

LTDA operates by learning relation-specific protein representations on the Lorentz manifold and then projecting them into tangent space. This adapter applies lightweight graph diffusion within the tangent space and decodes the diffused representations directly within this space. By utilizing the tangent space as a prediction-oriented interface, LTDA effectively bridges the gap between hierarchical representation learning and interaction classification.

Across SHS27K and SHS148K datasets under both BFS and DFS splits, LTDA consistently outperforms representative PPI methods and a reproduced HI-PPI baseline, demonstrating average gains of 1.41 percentage points in Micro-F1 score, 3.21 in AUPR, 1.31 in AUC, and 1.25 in accuracy. The improvements observed with LTDA cannot be attributed to simple smoothing or hub correction.

The study concludes that post-embedding adaptation, as implemented by LTDA, significantly enhances the effectiveness of Lorentz representations for PPI type prediction, highlighting the potential of this approach to address the challenges associated with incomplete and noisy protein-protein interaction networks.

Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at biorxiv.org →

More in AI

Why 'Local-First' Is the New Stack: How to Build Data-Sovereign AI Apps with Local LLMs, Private Vectors, and Zero-Cloud Dependencies

Originally published on tamiz.pro . For the past decade, the default architecture for software development has been heavily skewed toward the cloud.

  • Local-First architecture ensures data stays within user control.
  • Advances in local AI technologies enable LLMs to run on consumer hardware.
  • Practical implementation uses Python, Ollama, and ChromaDB for zero-cloud AI systems.

More from Sunday 4 October →