We introduce SegmA, a novel method for cryo-EM map visualization. By coloring voxels based on their amino acids, SegmA enhances both manual and automated modeling. This coloring approach also serves as a scoring function, extending its utility to de-novo structure modeling.SegmA’s algorithm is a cascade of group rotational equivariant convolutional neural networks (G-CNNs), enabling it to better handle the orientation variability of cryo-EM maps. During the training procedure of SegmA, an iterative co-training algorithm is utilized to filter out poorly labeled samples from the training dataset. SegmA also detects regions of low-confidence labeling, with amino acid detection accuracy increasing to 80% after removing these regions.Beyond 3D visualization, SegmA also has additional applications such as prediction of amino acid centers of mass, scoring of the fitting of a protein structural template into a cryo-EM map, and de-novo modeling of a protein complex.

Multiple sequence alignments (MSAs) are the workhorse of molecular evolution and structural biology research. From MSAs, the amino acids that are tolerated at each site during protein evolution can be inferred. However, little is known regarding the repertoire of tolerated amino acids in proteins when only a few or no sequence homologs are available, such as orphan and de novo designed proteins. Here we present EvoRator2, a deep-learning algorithm trained on over 15,000 protein structures that can predict which amino acids are tolerated at any given site, based exclusively on protein structural information mined from atomic coordinate files. We show that EvoRator2 obtained satisfying results for the prediction of position-weighted scoring matrices (PSSM). We further show that EvoRator2 obtained near state-of-the-art performance on proteins with high quality structures in predicting the effect of mutations in deep mutation scanning (DMS) experiments and that for certain DMS targets, EvoRator2 outperformed state-of-the-art methods. We also show that by combining EvoRator20s predictions with those obtained by a state-of-the-art deep-learning method that accounts for the information in the MSA, the prediction of the effect of mutation in DMS experiments was improved in terms of both accuracy and stability. EvoRator2 is designed to predict which amino-acid substitutions are tolerated in such proteins without many homologous sequences, including orphan or de novo designed proteins. We implemented our approach in the EvoRator web server (https://evorator.tau.ac.il).(c) 2023 Published by Elsevier Ltd.

Design of peptide binders is an attractive strategy for targeting “undruggable” protein-protein interfaces. Current design protocols rely on the extraction of an initial sequence from one known protein interactor of the target protein, followed by in-silico or in-vitro mutagenesis-based optimization of its binding affinity. Wet lab protocols can explore only a minor portion of the vast sequence space and cannot efficiently screen for other desirable properties such as high specificity and low toxicity, while in-silico design requires intensive computational resources and often relies on simplified binding models. Yet, for a multivalent protein target, dozens to hundreds of natural protein partners already exist in the cellular environment. Here, we describe a peptide design protocol that harnesses this diversity via a machine learning generative model. After identifying putative natural binding fragments by literature and homology search, a compositional Restricted Boltzmann Machine is trained and sampled to yield hundreds of diverse candidate peptides. The latter are further filtered via flexible molecular docking and an in-vitro microchip-based binding assay. We validate and test our protocol on calcineurin, a calcium-dependent protein phosphatase involved in various cellular pathways in health and disease. In a single screening round, we identified multiple 16-length peptides with up to six mutations from their closest natural sequence that successfully interfere with the binding of calcineurin to its substrates. In summary, integrating protein interaction and sequence databases, generative modeling, molecular docking and interaction assays enables the discovery of novel protein-protein interaction modulators.

Reliably scoring and ranking candidate models of protein complexes and assigning their oligomeric state from the structure of the crystal lattice represent outstanding challenges. A community-wide effort was launched to tackle these challenges. The latest resources on protein complexes and interfaces were exploited to derive a benchmark dataset consisting of 1677 homodimer protein crystal structures, including a balanced mix of physiological and non-physiological complexes. The non-physiological complexes in the benchmark were selected to bury a similar or larger interface area than their physiological counterparts, making it more difficult for scoring functions to differentiate between them. Next, 252 functions for scoring protein-protein interfaces previously developed by 13 groups were collected and evaluated for their ability to discriminate between physiological and non-physiological complexes. A simple consensus score generated using the best performing score of each of the 13 groups, and a cross-validated Random Forest (RF) classifier were created. Both approaches showed excellent performance, with an area under the Receiver Operating Characteristic (ROC) curve of 0.93 and 0.94, respectively, outperforming individual scores developed by different groups. Additionally, AlphaFold2 engines recalled the physiological dimers with significantly higher accuracy than the non-physiological set, lending support to the reliability of our benchmark dataset annotations. Optimizing the combined power of interface scoring functions and evaluating it on challenging benchmark datasets appears to be a promising strategy.

Predicting the functional sites of a protein from its structure, such as the binding sites of small molecules, other proteins or antibodies sheds light on its function in vivo. Currently, two classes of methods prevail: Machine Learning (ML) models built on top of handcrafted features and comparative modeling. They are respectively limited by the expressivity of the handcrafted features and the availability of similar proteins. Here, we introduce ScanNet, an end-to-end, interpretable geometric deep learning model that learns features directly from 3D structures. ScanNet builds representations of atoms and amino acids based on the spatio-chemical arrangement of their neighbors. We train ScanNet for detecting protein-protein and protein-antibody binding sites, demonstrate its accuracy - including for unseen protein folds - and interpret the filters learned. Finally, we predict epitopes of the SARS-CoV-2 spike protein, validating known antigenic regions and predicting previously uncharacterized ones. Overall, ScanNet is a versatile, powerful, and interpretable model suitable for functional site prediction tasks. A webserver for ScanNet is available from http://bioinfo3d.cs.tau.ac.il/ScanNet/

The SARS-CoV-2 Omicron variant evades most neutralizing vaccine-induced antibodies and is associated with lower antibody titers upon breakthrough infections than previous variants. However, the mechanism remains unclear. Here, we find using a geometric deep-learning model that Omicron's extensively mutated receptor binding site (RBS) features reduced antigenicity compared with previous variants. Mice immunization experiments with different recombinant receptor binding domain (RBD) variants confirm that the serological response to Omicron is drastically attenuated and less potent. Analyses of serum cross-reactivity and competitive ELISA reveal a reduction in antibody response across both variable and conserved RBD epitopes. Computational modeling confirms that the RBS has a potential for further antigenicity reduction while retaining efficient receptor binding. Finally, we find a similar trend of antigenicity reduction over decades for hCoV229E, a common cold coronavirus. Thus, our study explains the reduced antibody titers associated with Omicron infection and reveals a possible trajectory of future viral evolution.

Designing peptides for protein–protein interaction inhibition is of significant interest in computer‐aided drug design. Such inhibitory peptides could mimic and compete with the binding of the partner protein to the inhibition target. Experimental peptide design is a laborious, time consuming, and expensive multi‐step process. Therefore, in silico peptide design can be beneficial for achieving this task. We present a novel algorithm, Pep–Whisperer, which aims to design inhibitory peptides for protein–protein interaction. The desirable peptides would have a relatively high predicted binding affinity to the target protein in a given protein–protein complex. The algorithm outputs linear peptides which are based on an initial template. The template could either be a peptide which is retrieved from the interaction site, or a patch of nonconsecutive amino acids from the protein–protein interface which is completed to a linear peptide by short polyalanine linkers. In addition, the algorithm takes into consideration the conservation of the amino acids in the ligand‐protein binding site by using evolutionary information for choosing the preferred amino acids in each position of the designed peptide. Our algorithm was able to design peptides with high predicted binding affinity to the target protein. The method is fully automated and available as a web server at http://bioinfo3d.cs.tau.ac.il/PepWhisperer/ .

Predicting the various binding sites of a protein from its structure sheds light on its function and paves the way towards design of interaction inhibitors. Here, we report ScanNet, a freely available web server for prediction of protein-protein, protein - disordered protein and protein - antibody binding sites from structure. ScanNet (Spatio-Chemical Arrangement of Neighbors Network) is an end-to-end, interpretable geometric deep learning model that learns spatio-chemical patterns directly from 3D structures. ScanNet consistently outperforms Machine Learning models based on handcrafted features and comparative modeling approaches. The web server is linked to both the PDB and AlphaFoldDB, and supports user-provided structure files. Predictions can be readily visualized on the website via the Molstar web app and locally via ChimeraX. ScanNet is available at http://bioinfo3d.cs.tau.ac.il/ScanNet/.

The molecular chaperone Hsp90 is a ubiquitous ATPase-directed protein responsible for the activation and structural stabilization of a large clientele of proteins. As such, Hsp90 has emerged as a suitable candidate for the treatment of a diverse set of diseases, such as cancer and neurodegeneration. The inhibition of the chaperone through ATP-competitive inhibitors, however, was shown to lead to undesirable side effects. One strategy to alleviate this problem is the development of molecules that are able to disrupt specific protein-protein interactions, thus modulating the activity of Hsp90 only in the particular cellular pathway that needs to be targeted. Here, we exploit novel computational and theoretical approaches to design a set of peptides that are able to bind Hsp90 and compete for its interaction with the co-chaperone Cdc37, which is found to be responsible for the promotion of cancer cell proliferation. In spite of their capability to disrupt the Hsp90-Cdc37 interaction, no important cytotoxicity was observed in human cancer cells exposed to designed compounds. These findings imply the need for further optimization of the compounds, which may lead to new ways of interfering with the Hsp90 mechanisms that are important for tumour growth.

Macromolecular complexes play a key role in cellular function. Predicting the structure and dynamics of these complexes is one of the key challenges in structural biology. Docking applications have traditionally been used to predict pairwise interactions between proteins. However, few methods exist for modeling multi-protein assemblies. Here we present two methods, CombDock and DockStar, that can predict multi-protein assemblies starting from subunit structural models. CombDock can assemble subunits without any assumptions about the pairwise interactions between subunits, while DockStar relies on the interaction graph or, alternatively, a homology model or a cryo-electron microscopy (EM) density map of the entire complex. We demonstrate the two methods using RNA polymerase II with 12 subunits and TRiC/CCT chaperonin with 16 subunits.

The recent cryo-EM resolution revolution enables the development of algorithms for direct de-novo modeling of protein structures into cryo-EM density maps. Here we present a machine learning based method for the detection of high confidence anchor amino acid residues in such a map. Such anchor residues can be exploited in several local de-novo modeling tasks, such as the reliable positioning of secondary structures, loop modeling and general fragment based modeling. In the experimental results we show the ability of the proposed procedure to locate and classify a significant number of amino acids in density maps of 3.1 Å (or better) resolution. Our performance analysis indicates that the main factor affecting the detection accuracy is the lack of sufficient experimental data for the training stage of the algorithm. Thus, our method is expected to improve significantly in the near future, due to the rapid increase in the release of novel high resolution cryo-EM maps.

In this chapter we present two methods related to rational design of inhibitory peptides: PepCrawler: A tool to derive binding peptides from protein-protein complexes and the prediction of protein-peptide complexes. Given an initial protein-peptide complex, the method detects improved predicted peptide binding conformations which bind the protein with higher affinity. This program is a robotics motivated algorithm, representing the peptide as a robotic arm moving among obstacles and exploring its conformational space in an efficient way. PinaColada: A peptide design program for the discovery of novel peptide candidates that inhibit protein-protein interactions. PinaColada uses PepCrawler while introducing sequence mutations, in order to find novel inhibitory peptides for PPIs. It uses the ant colony optimization approach to explore the peptide's sequence space, while using PepCrawler in the refinement stage.

Motivation: A highly efficient template-based protein-protein docking algorithm, nicknamed SnapDock, is presented. It employs a Geometric Hashing-based structural alignment scheme to align the target proteins to the interfaces of non-redundant protein-protein interface libraries. Docking of a pair of proteins utilizing the 22 600 interface PIFACE library is performed in < 2 min on the average. A flexible version of the algorithm allowing hinge motion in one of the proteins is presented as well.Results: To evaluate the performance of the algorithm a blind re-modelling of 3547 PDB complexes, which have been uploaded after the PIFACE publication has been performed with success ratio of about 35%. Interestingly, a similar experiment with the template free PatchDock docking algorithm yielded a success rate of about 23% with roughly 1/3 of the solutions different from those of SnapDock. Consequently, the combination of the two methods gave a 42% success ratio.

MOTIVATION:Design of protein-protein interaction (PPI) inhibitors is a major challenge in Structural Bioinformatics. Peptides, especially short ones (5-15 amino acid long), are natural candidates for inhibition of protein-protein complexes due to several attractive features such as high structural compatibility with the protein binding site (mimicking the surface of one of the proteins), small size and the ability to form strong hotspot binding connections with the protein surface. Efficient rational peptide design is still a major challenge in computer aided drug design, due to the huge space of possible sequences, which is exponential in the length of the peptide, and the high flexibility of peptide conformations.RESULTS:In this article we present PinaColada, a novel computational method for the design of peptide inhibitors for protein-protein interactions. We employ a version of the ant colony optimization heuristic, which is used to explore the exponential space ([Formula: see text]) of length n peptide sequences, in combination with our fast robotics motivated PepCrawler algorithm, which explores the conformational space for each candidate sequence. PinaColada is being run in parallel, on a DELL PowerEdge 2.8 GHZ computer with 20 cores and 256 GB memory, and takes up to 24 h to design a peptide of 5-15 amino acids length.AVAILABILITY AND IMPLEMENTATION:An online server available at: http://bioinfo3d.cs.tau.ac.il/PinaColada/.CONTACT:danielza@post.tau.ac.il; wolfson@tau.ac.il.

MOTIVATIONA wide range of fundamental biological processes are mediated by membrane proteins. Despite their large number and importance, less than 1% of all 3D protein structures deposited in the Protein Data Bank are of membrane proteins. This is mainly due to the challenges of crystallizing such proteins or performing NMR spectroscopy analyses. All the more so, there is only a small number of membrane protein-protein complexes with known structure. Therefore, developing computational tools for docking membrane proteins is crucial. Numerous methods for docking globular proteins exist, however few have been developed especially for membrane proteins and designed to address docking within the lipid bilayer environment.RESULTSWe present a novel algorithm, Memdock, for docking α-helical membrane proteins which takes into consideration the lipid bilayer environment for docking as well as for refining and ranking the docking candidates. We show that our algorithm improves both the docking accuracy and the candidates ranking compared to a standard protein-protein docking algorithm.AVAILABILITY AND IMPLEMENTATIONhttp://bioinfo3d.cs.tau.ac.il/Memdock/CONTACTSnamih@tau.ac.il or wolfson@tau.ac.ilSUPPLEMENTARY INFORMATIONSupplementary data are available at Bioinformatics online.

A major challenge in protein-protein docking is the distinction between near-native and decoy complex predictions. It has been shown that near native solutions are usually located at the bottom of deep and densely populated funnels in the binding energy plot of the complex. Thus exploration, whether the energy plot of the vicinity of a docking solution is "funnel like", can serve as a validation of such a solution. Generation of such densely sampled plots, however, is a major computational challenge. We have designed an accurate and highly efficient parallel algorithm for generation of such energy plots and implemented it on a server with 4 GPU processors, each with 2880 cores. The algorithm achieved a speedup of about 150 compared to its serial counterpart, while even outperforming it in the achieved results. While the algorithm proved very useful for near native complex hypothesis validation, it still detects many funnels for decoy solutions, especially those with good shape complementarity.

Motivation: Atomic resolution modeling of large multimolecular assemblies is a key task in Structural Cell Biology. Experimental techniques can provide atomic resolution structures of single proteins and small complexes, or low resolution data of large multimolecular complexes. Results: We present a novel integrative computational modeling method, which integrates both low and high resolution experimental data. The algorithm accepts as input atomic resolution structures of the individual subunits obtained from X-ray, NMR or homology modeling, and interaction data between the subunits obtained from mass spectrometry. The optimal assembly of the individual subunits is formulated as an Integer Linear Programming task. The method was tested on several representative complexes, both in the bound and unbound cases. It placed correctly most of the subunits of multimolecular complexes of up to 16 subunits and significantly outperformed the CombDock and Haddock multimolecular docking methods. Availability and implementation: http://bioinfo3d.cs.tau.ac.il/DockStar Contact: naamaamir@mail.tau.ac.il or wolfson@tau.ac.il Supplementary information: Supplementary data are available at Bioinformatics online.

LIM kinases (LIMKs) are important cell cytoskeleton regulators that play a prominent role in cancer manifestation and neuronal diseases. The LIMK family consists of two homologues, LIMK1 and LIMK2, which differ from one another in expression profile, intercellular localization, and function. The main substrate of LIMK is cofilin, a member of the actin-depolymerizing factor (ADF) protein family. When phosphorylated by LIMK, cofilin is inactive. LIMKs play a contributory role in several neurodevelopmental disorders and in cancer growth and metastasis. We recently reported the development and validation of a novel LIMK inhibitor, referred to here as T56-LIMKi, using a combination of computational methods and classical biochemistry techniques. Here we report that T56-LIMKi inhibits LIMK2 with high specificity, and shows little or no cross-reactivity with LIMK1. We found that T56-LIMKi decreases phosphorylated cofilin (p-cofilin) levels and thus inhibits growth of several cancerous cell lines, including those of pancreatic cancer, glioma and schwannoma. Because the most promising in-vitro effect of T56-LIMKi was observed in the pancreatic cancer cell line Panc-1, we tested the inhibitor on a nude mouse Panc-1 xenograft model. T56-LIMKi reduced tumor size and p-cofilin levels in the Panc-1 tumors, leading us to propose T56-LIMKi as a candidate drug for cancer therapy.