Protein complex prediction with AlphaFold-Multimer

While the vast majority of well-structured single protein chains can now be predicted to high accuracy due to the recent AlphaFold [1] model, the prediction of multi-chain protein complexes remains a challenge in many cases. In this work, we demonstrate that an AlphaFold model trained specifically for multimeric inputs of known stoichiometry, which we call AlphaFold-Multimer, significantly increases accuracy of predicted multimeric interfaces over input-adapted single-chain AlphaFold while maintaining high intra-chain accuracy. On a benchmark dataset of 17 heterodimer proteins without templates (introduced in [2]) we achieve at least medium accuracy (DockQ [3]≥0.49) on 14 targets and high accuracy (DockQ≥0.8) on 6 targets, compared to 9 targets of at least medium accuracy and 4 of high accuracy for the previous state of the art system (an AlphaFold-based system from [2]). We also predict structures for a large dataset of 4,433 recent protein complexes, from which we score all non-redundant interfaces with low template identity. For heteromeric interfaces we successfully predict the interface (DockQ≥0.23) in 67% of cases, and produce high accuracy predictions (DockQ≥0.8) in 23% of cases, an improvement of +25 and +11 percentage points over the flexible linker modification of AlphaFold [4] respectively. For homomeric interfaces we successfully predict the interface in 69% of cases, and produce high accuracy predictions in 34% of cases, an improvement of +5 percentage points in both instances.

Hierarchical Hidden Community Detection for Protein Complex Prediction

BackgroundDiscovering functional modules in protein-protein interaction networks through optimization remains a longstanding challenge in Biology. Traditional algorithms simply consider strong protein complexes found in the original network by optimizing some metric, which may cause obstacles for discovering weak and hidden complexes that are overshadowed by strong complexes. Additionally, protein complexes have not only different densities but also various ranges of scales, making them extremely difficult to be detected. We address these issues and propose a hierarchical hidden community detection approach to predict protein complexes of various strengths and scales accurately. ResultsWe propose a meta-method called HirHide (Hierarchical Hidden Community Detection). It is the first combination of hierarchical structure with hidden structure, which provides a new perspective for finding protein complexes of various strengths and scales. We compare the performance of several standard community detection methods with their HirHide versions. Experimental results show that the HirHide versions achieve better performance and sometimes even significantly outperform the baselines. ConclusionsHirHide can adopt any standard community detection method as the base algorithm and enable it to discover hidden hierarchical communities as well as boosting the detection of strong hierarchical communities. Some biological networks are too complex for standard community detection algorithms to produce a positive performance. Most of the time, a better choice is to choose a corresponding algorithm based on the characteristics of a specific biological network. Under these circumstances, HirHide has clear advantages because of its flexibility. At the same time, according to the natural hierarchy of cells, organelle, intracellular compound etc., hierarchical structure with hidden structure is in line with the characteristics of the data itself, thus helping researchers to study biological interactions more deeply.

Protein complex similarity based on Weisfeiler-Lehman labeling

Being able to quantify the similarity between two protein complexes is essential for numerous applications. Prominent examples are database searches for known complexes with a given query complex, comparison of the output of different protein complex prediction algorithms, or summarizing and clustering protein complexes, e.g., for visualization. While the corresponding problems have received much attention on single proteins and protein families, the question about how to model and compute similarity between protein complexes has not yet been systematically studied. Because protein complexes can be naturally modeled as graphs, in principle general graph similarity measures may be used, but these are often computationally hard to obtain and do not take typical properties of protein complexes into account. Here we propose a parametric family of similarity measures based on Weisfeiler-Lehman labeling. We evaluate it on simulated complexes of the extended human integrin adhesome network. Because the connectivity (graph topology) of real complexes is often unknown and hard to obtain experimentally, we use both known protein-protein interaction networks and known interdependencies (constraints) between interactions to simulate more realistic complexes than from interaction networks alone. We empirically show that the defined family of similarity measures is in good agreement with edit similarity, a similarity measure derived from graph edit distance, but can be much more efficiently computed. It can therefore be used in large-scale studies and simulations and serve as a basis for further refinements of modeling protein complex similarity.

Protein complex similarity based on Weisfeiler-Lehman labeling

Being able to quantify the similarity between two protein complexes is essential for numerous applications. Prominent examples are database searches for known complexes with a given query complex, comparison of the output of different protein complex prediction algorithms, or summarizing and clustering protein complexes, e.g., for visualization. While the corresponding problems have received much attention on single proteins and protein families, the question about how to model and compute similarity between protein complexes has not yet been systematically studied. Because protein complexes can be naturally modeled as graphs, in principle general graph similarity measures may be used, but these are often computationally hard to obtain and do not take typical properties of protein complexes into account. Here we propose a parametric family of similarity measures based on Weisfeiler-Lehman labeling. We evaluate it on simulated complexes of the extended human integrin adhesome network. Because the connectivity (graph topology) of real complexes is often unknown and hard to obtain experimentally, we use both known protein-protein interaction networks and known interdependencies (constraints) between interactions to simulate more realistic complexes than from interaction networks alone. We empirically show that the defined family of similarity measures is in good agreement with edit similarity, a similarity measure derived from graph edit distance, but can be much more efficiently computed. It can therefore be used in large-scale studies and simulations and serve as a basis for further refinements of modeling protein complex similarity.