Welcome to ELIXIR Czech Meeting

From studying protein dynamics to unveiling cryptic binding sites or assessing the effectiveness of ligand binding site prediction software, access to several snapshots of a protein is needed. Availability of both bound (holo) and unbound (apo) forms of a protein is paramount for making meaningful comparisons and drawing robust conclusions. The few existing resources that provide access to such data are restricted either in terms of protein coverage, or in the number of provided structure pairs which does not always reflect the conformational variance that is represented by the structures deposited in the Protein Data Bank (PDB).

Here we present previously designed application (AHoJ, Apo-Holo Juxtaposition) and use it to perform an extensive search for apo-holo pairs for each individual protein-ligand interaction across the PDB (~500,000 small molecule interactions, excluding interactions with peptides and nucleic acids). We assemble the results of this search into a database that can be used to train and evaluate predictors, discover potentially druggable proteins, and reveal associations that can confirm existing hypotheses or expose protein- and ligand-specific relationships like order-to-disorder transitions, that were previously obscured by intermittent or partial data.

The accurate prediction of protein-ligand binding sites is crucial for understanding protein interactions, especially in the context of biotechnology and drug discovery. There are two main approaches to tackle this challenge: one relies on the sequence of the protein (sequence-based methods), while the other relies on the three-dimensional structure of the protein (structure-based methods).

In this talk, we will discuss a novel approach that combines the strengths of both approaches to advance the state-of-the-art in this field. Our hybrid model merges two cutting-edge deep learning techniques: protein language models (pLMs) from the sequence-based approach and Graph Neural Networks (GNNs) from the structure-based approach. Specifically, we create a residue-level Graph Attention Network (GAT) model using the 3D protein structure and incorporate pre-trained pLM embeddings as node features. This integration allows our model to capture both the sequential information embedded in the protein sequence and the structural relationships within the protein.

Our model performs well compared to existing methods on a benchmark dataset, covering various ligands and ligand types. Ablation studies highlight the importance of the graph attention mechanism, particularly in densely connected graphs. Furthermore, we illustrate that as we employ more intricate pLMs to represent node features, the relative impact of the GNN architecture diminishes. This finding suggests that, to some extent, the structural information necessary for accurate binding site prediction is inherently encoded within the pLMs themselves.

Thanks to advanced structural biology approaches, more than 200,000 experimentally determined protein structures are available in the Protein Data Bank. Based on this data, more than 200,000,000 protein structures were generated using artificial intelligence algorithms and are available in AlphaFoldDB. This data has greatly expanded the possibilities of studying protein families, their anatomy, variability, common features, and evolutionary conservation. Visualizing different aspects of protein family structures provides important information for their analysis and research.

In this paper, we would like to present new methodologies for visualizing protein families and their properties. Specifically, 1D diagrams of protein families, generated using the OverProt tool [1], 2D diagrams, produced by the 2DProts application [2], and mapping properties and annotations of protein families onto these diagrams. The properties include, e.g., partial atomic charges calculated using the software tools ACC II [3] and αCharges [4].

1. A. Midlik, I. Hutařová Vařeková, J. Hutař, A. Chareshneu, K. Berka, R. Svobodová, Bioinf., 38(14), (2022), 3648-3650. 2. 2. I, Hutařová Vařeková, J. Hutař, A. Midlik, V. Horský, E. Hladká, R. Svobodová, K. Berka, Bioinf., 37(23), (2021), 4599-4601. 3. T. Raček, O. Schindler, D. Toušek, V. Horský, K. Berka, J. Koča, R. Svobodová, Nucl. Acids Res., 48(W1), (2020), W591-W596. 4. O. Schindler, K. Berka, A. Cantara, A. Křenek, D. Tichý, T. Raček, R. Svobodová, Nucl. Acids Res., 51(W1), (2023), W11–W16.

Core Facility Biological Data Management and Analysis of CEITEC Masaryk University, supported by ELIXIR CZ research infrastructure (MEYS Grant No: LM2023055) is gratefully acknowledged for the obtaining of the scientific data presented in this contribution.