Predicting multicellular function through multi-layer tissue networks (1707.04638v1)

Abstract: Motivation: Understanding functions of proteins in specific human tissues is essential for insights into disease diagnostics and therapeutics, yet prediction of tissue-specific cellular function remains a critical challenge for biomedicine. Results: Here we present OhmNet, a hierarchy-aware unsupervised node feature learning approach for multi-layer networks. We build a multi-layer network, where each layer represents molecular interactions in a different human tissue. OhmNet then automatically learns a mapping of proteins, represented as nodes, to a neural embedding based low-dimensional space of features. OhmNet encourages sharing of similar features among proteins with similar network neighborhoods and among proteins activated in similar tissues. The algorithm generalizes prior work, which generally ignores relationships between tissues, by modeling tissue organization with a rich multiscale tissue hierarchy. We use OhmNet to study multicellular function in a multi-layer protein interaction network of 107 human tissues. In 48 tissues with known tissue-specific cellular functions, OhmNet provides more accurate predictions of cellular function than alternative approaches, and also generates more accurate hypotheses about tissue-specific protein actions. We show that taking into account the tissue hierarchy leads to improved predictive power. Remarkably, we also demonstrate that it is possible to leverage the tissue hierarchy in order to effectively transfer cellular functions to a functionally uncharacterized tissue. Overall, OhmNet moves from flat networks to multiscale models able to predict a range of phenotypes spanning cellular subsystems

• The paper introduces a hierarchy-aware unsupervised algorithm to predict tissue-specific protein functions via multi-layer tissue networks.
• It employs a neural embedding to capture protein similarities across 107 tissues, outperforming traditional models.
• The model enables effective function transfer predictions in 48 tissues, surpassing state-of-the-art techniques by up to 20.3% in transfer learning.

Predicting Multicellular Function Through Multi-Layer Tissue Networks

The paper by Zitnik and Leskovec addresses a significant challenge in biomedicine: predicting tissue-specific protein functions. The authors introduce a hierarchy-aware unsupervised learning algorithm designed for multi-layer networks to uncover these functions. Each network layer corresponds to molecular interactions within a different human tissue, allowing the model to capture the diversity of cellular functions across different tissue contexts.

Summary of Methodology

The core contribution of this research lies in representing each tissue as a network layer while leveraging a hierarchical model that defines relationships between these layers. This multi-layered approach enables the extraction of protein features via a neural embedding into a low-dimensional space. A significant focus is placed on maintaining feature similarities among proteins sharing similar network neighborhoods and tissues. The methodology builds on advancements in unsupervised representation learning and structured regularization to capture the complex tissue organizations that traditional methods often neglect.

Experimental Evaluation

The model was evaluated on a multiplex network of 107 human tissues, demonstrating its superior capabilities over existing techniques. In tissue-specific function prediction across 48 tissues, it showed enhanced predictive accuracy. The hierarchical approach was instrumental in allowing function transfer predictions in uncharacterized tissues, with the method outperforming state-of-the-art techniques with margins up to 14.9% for classification tasks and 20.3% for transfer learning.

Implications and Future Directions

This research has several implications:

1. Biological Insightfulness: By shifting from flat to multiscale models, the approach provides deeper insights into the phenotypic manifestations of cellular subsystems, enabling more precise therapeutic strategies and disease diagnostics.
2. Advances in Feature Learning: The introduction of hierarchy-aware structured regularization marks an important advance in handling multi-scale dependencies for feature learning, showing potential applicability beyond tissue networks.
3. Transfer Learning Capabilities: The ability to transfer cellular functions across tissues highlights the model’s utility in addressing knowledge gaps in unexplored biological contexts.

Future research could focus on extending the model to handle graph-based dependencies, reflecting real-world scenarios of more complex tissue interdependencies. Furthermore, applications could move beyond cellular functions to areas such as gene-disease associations and pathway detections.

In conclusion, this paper provides a substantial step forward in computational biomedicine by enhancing the precision of tissue-specific protein functional predictions while paving the way for future explorations in hierarchical multi-layer learning models.