Growing ecosystem of deep learning methods for modeling protein$\unicode{x2013}$protein interactions

Julia R. Rogers,Gergő Nikolényi,Mohammed AlQuraishi
2023-12-07
Abstract:Numerous cellular functions rely on protein$\unicode{x2013}$protein interactions. Efforts to comprehensively characterize them remain challenged however by the diversity of molecular recognition mechanisms employed within the proteome. Deep learning has emerged as a promising approach for tackling this problem by exploiting both experimental data and basic biophysical knowledge about protein interactions. Here, we review the growing ecosystem of deep learning methods for modeling protein interactions, highlighting the diversity of these biophysically-informed models and their respective trade-offs. We discuss recent successes in using representation learning to capture complex features pertinent to predicting protein interactions and interaction sites, geometric deep learning to reason over protein structures and predict complex structures, and generative modeling to design de novo protein assemblies. We also outline some of the outstanding challenges and promising new directions. Opportunities abound to discover novel interactions, elucidate their physical mechanisms, and engineer binders to modulate their functions using deep learning and, ultimately, unravel how protein interactions orchestrate complex cellular behaviors.
Biomolecules,Machine Learning
What problem does this paper attempt to address?
This paper reviews the application of deep learning in modeling protein-protein interactions. The study points out that comprehensive understanding of these interactions is crucial for studying the molecular recognition mechanisms due to their diversity. Despite the availability of various methods for detecting and characterizing these interactions, challenges such as data sparsity, sensitivity, and specificity still exist. Deep learning offers new opportunities to predict and design protein interactions by leveraging experimental data and biophysical knowledge. The paper mentions several deep learning techniques such as representation learning (for capturing complex predictive interactions and features of binding sites), geometric deep learning (for handling protein structures and predicting complex formation structures), and generative modeling (for designing new protein assemblies). Additionally, the paper discusses current challenges such as lack of generalization ability and future possible directions, including expanding the applicability of models to cover transient and disordered interactions. In summary, this paper aims to address how deep learning can be used to better understand and simulate the complexity of protein-protein interactions, thereby revealing the mechanisms of cellular behavior and providing new avenues for disease treatment and the design of protein regulators.