Modeling interactions between proteins has been crucial for uncovering the structure, function, and therapeutic potential of proteins. Extensive efforts to develop experimental and computational technologies to construct and analyze protein interaction networks have improved the characterization of proteins. However, protein interaction networks are typically presented as generic maps without contextual information about tissues or celltypes. Despite the development of high-throughput methods for screening protein-protein interactions and sequencing technologies to measure gene expression with single-cell resolution, accurately modeling protein interactions across diverse biological contexts, such as tissues and celltypes, remains a critical experimental and computational challenge.

The roles of proteins is influenced by biological contexts in which they are found:

• While nearly every cell contains the same genome, the expression and function of a protein depends on the cell or tissue. Further, protein expression and function can differ significantly between healthy and diseased cells/tissues.
• The ability to model and interrogate proteins in diverse biological contexts can improve the characterization of a disease’s mechanism of action and the design of safe and efficacious drugs.

There is a growing need to develop methodologies that can effectively inject and leverage contextual information of proteins. Still, existing algorithms are limited in their capacity to model proteins with celltype or tissue specificity.

Overview of AWARE

AWARE is a self-supervised geometric deep learning model that can generate protein representations in different celltype contexts. AWARE integrates single cell transcriptomics data with a protein interaction network, celltype interaction network, and tissue hierarchy to generate protein representations with celltype resolution.

In this work, we focus on protein-coding genes and do not encode differences of protein isoforms (e.g., due to alternative splicing). Unlike existing methods that provide only one representation per protein (assuming each protein-coding gene encodes only one protein), resulting in fewer than 22,000 protein representations, AWARE generates a unique representation for each celltype that a protein is activated in.

With our 394,760 contextualized protein representations (i.e., protein representations injected with celltype-specificity), we demonstrate AWARE’s ability to integrate structured and transcriptomic data, perform transfer learning across proteins, celltypes, and tissues, and generate contextualized predictions for diverse biomedical tasks.

AWARE Algorithm

AWARE is a self-supervised geometric deep learning model that can generate protein representations in diverse celltype contexts. AWARE is trained on a set of celltype specific protein interaction networks unified by a cellular and tissue network to produce contextualized protein representations based celltype activation. Unlike existing approaches, which do not consider biological context, AWARE produces multiple representations of proteins based on context, representations of the celltypes themselves, and representations of the tissues from which the celltypes are derived and the tissue hierarchy.

Given the multi-scale nature of the model inputs, AWARE is equipped to learn protein-level, celltype-level, and tissue-level topology in a single unified embedding space. To fully leverage the multi-scale inputs, AWARE uses protein-, celltype-, and tissue-level attention mechanisms and objective functions to inject cellular and tissue organization into the embedding space. AWARE is designed such that pairs of nodes that share an edge are embedded nearby each other, protein representations of the same celltype are embedded close by (and far from protein representations of a different celltype), and protein representations are embedded close to the representation of their corresponding celltype (and far from other celltype representations).

AWARE propagates message on proteins, celltypes, and tissues using attention mechanisms specific to each node and relationship type:

1. The protein-level objective function, which considers self-supervised link prediction on the protein interactions and celltype-identity classification on the protein nodes, enables AWARE to produce an embedding space that captures both the topology of the celltype-specific protein interaction networks and the celltype identity of proteins.
2. The celltype- and tissue-specific components in celltype- and tissue-specific objective functions are based solely on self-supervised link prediction to learn cellular and tissue organization.
3. Such information is propagated to the protein representations using an attention bridge, imposing tissue and cellular organization to the protein representations.

Publication

Contextualizing Protein Representations Using Deep Learning on Interactomes and Single-Cell Experiments
Michelle M. Li, Yepeng Huang, Marissa Sumathipala, Man Qing Liang, Alberto Valdeolivas, Ashwin N. Ananthakrishnan, Katherine Liao, Daniel Marbach and Marinka Zitnik
In Review 2023

@article{li2023contextualizing,
  title={Contextualizing Protein Representations Using Deep Learning on Interactomes and Single-Cell Experiments},
  author={Li, Michelle M and Huang, Yepeng and Sumathipala, Marissa and Liang, Man Qing and Valdeolivas, Alberto and Ananthakrishnan, Ashwin N and Marbach, Daniel and Zitnik, Marinka},
  journal={bioRxiv},
  url={},
  year={2023}
}

Code Availability

Pytorch implementation of AWARE is available in the GitHub repository.

We provide an interactive demo to explore AWARE’s protein representations through a visual interface in the HuggingFace Space.

Authors

• Michelle M. Li
• Yepeng Huang
• Marissa Sumathipala
• Man Qing Liang
• Alberto Valdeolivas
• Katherine Liao
• Daniel Marbach
• Marinka Zitnik