AlphaFold2-multimer guided high-accuracy prediction of typical and atypical ATG8-binding motifs.

Macroautophagy/autophagy is an intracellular degradation process central to cellular homeostasis and defense against pathogens in eukaryotic cells. Regulation of autophagy relies on hierarchical binding of autophagy cargo receptors and adaptors to ATG8/LC3 protein family members. Interactions with ATG8/LC3 are typically facilitated by a conserved, short linear sequence, referred to as the ATG8/LC3 interacting motif/region (AIM/LIR), present in autophagy adaptors and receptors as well as pathogen virulence factors targeting host autophagy machinery. Since the canonical AIM/LIR sequence can be found in many proteins, identifying functional AIM/LIR motifs has proven challenging. Here, we show that protein modelling using Alphafold-Multimer (AF2-multimer) identifies both canonical and atypical AIM/LIR motifs with a high level of accuracy. AF2-multimer can be modified to detect additional functional AIM/LIR motifs by using protein sequences with mutations in primary AIM/LIR residues. By combining protein modelling data from AF2-multimer with phylogenetic analysis of protein sequences and protein-protein interaction assays, we demonstrate that AF2-multimer predicts the physiologically relevant AIM motif in the ATG8-interacting protein 2 (ATI-2) as well as the previously uncharacterized noncanonical AIM motif in ATG3 from potato (Solanum tuberosum). AF2-multimer also identified the AIM/LIR motifs in pathogen-encoded virulence factors that target ATG8 members in their plant and human hosts, revealing that cross-kingdom ATG8-LIR/AIM associations can also be predicted by AF2-multimer. We conclude that the AF2-guided discovery of autophagy adaptors/receptors will substantially accelerate our understanding of the molecular basis of autophagy in all biological kingdoms.

Modeling Flexible Protein Structure With AlphaFold2 and Crosslinking Mass Spectrometry.

We propose a pipeline that combines AlphaFold2 (AF2) and crosslinking mass spectrometry (XL-MS) to model the structure of proteins with multiple conformations. The pipeline consists of two main steps: ensemble generation using AF2 and conformer selection using XL-MS data. For conformer selection, we developed two scores-the monolink probability score (MP) and the crosslink probability score (XLP)-both of which are based on residue depth from the protein surface. We benchmarked MP and XLP on a large dataset of decoy protein structures and showed that our scores outperform previously developed scores. We then tested our methodology on three proteins having an open and closed conformation in the Protein Data Bank: Complement component 3 (C3), luciferase, and glutamine-binding periplasmic protein, first generating ensembles using AF2, which were then screened for the open and closed conformations using experimental XL-MS data. In five out of six cases, the most accurate model within the AF2 ensembles-or a conformation within 1 Å of this model-was identified using crosslinks, as assessed through the XLP score. In the remaining case, only the monolinks (assessed through the MP score) successfully identified the open conformation of glutamine-binding periplasmic protein, and these results were further improved by including the "occupancy" of the monolinks. This serves as a compelling proof-of-concept for the effectiveness of monolinks. In contrast, the AF2 assessment score was only able to identify the most accurate conformation in two out of six cases. Our results highlight the complementarity of AF2 with experimental methods like XL-MS, with the MP and XLP scores providing reliable metrics to assess the quality of the predicted models. The MP and XLP scoring functions mentioned above are available at https://gitlab.com/topf-lab/xlms-tools.

Combined prediction and design reveals the target recognition mechanism of an intrinsically disordered protein interaction domain.

An increasing number of protein interaction domains and their targets are being found to be intrinsically disordered proteins (IDPs). The corresponding target recognition mechanisms are mostly elusive because of challenges in performing detailed structural analysis of highly dynamic IDP-IDP complexes. Here, we show that by combining recently developed computational approaches with experiments, the structure of the complex between the intrinsically disordered C-terminal domain (CTD) of protein 4.1G and its target IDP region in NuMA can be dissected at high resolution. First, we carry out systematic mutational scanning using dihydrofolate reductase-based protein complementarity analysis to identify essential interaction regions and key residues. The results are found to be highly consistent with an α/ß-type complex structure predicted by AlphaFold2 (AF2). We then design mutants based on the predicted structure using a deep learning protein sequence design method. The solved crystal structure of one mutant presents the same core structure as predicted by AF2. Further computational prediction and experimental assessment indicate that the well-defined core structure is conserved across complexes of 4.1G CTD with other potential targets. Thus, we reveal that an intrinsically disordered protein interaction domain uses an α/ß-type structure module formed through synergistic folding to recognize broad IDP targets. Moreover, we show that computational prediction and experiment can be jointly applied to segregate true IDP regions from the core structural domains of IDP-IDP complexes and to uncover the structure-dependent mechanisms of some otherwise elusive IDP-IDP interactions.

Improved structure-related prediction for insufficient homologous proteins using MSA enhancement and pre-trained language model.

In recent years, protein structure problems have become a hotspot for understanding protein folding and function mechanisms. It has been observed that most of the protein structure works rely on and benefit from co-evolutionary information obtained by multiple sequence alignment (MSA). As an example, AlphaFold2 (AF2) is a typical MSA-based protein structure tool which is famous for its high accuracy. As a consequence, these MSA-based methods are limited by the quality of the MSAs. Especially for orphan proteins that have no homologous sequence, AlphaFold2 performs unsatisfactorily as MSA depth decreases, which may pose a barrier to its widespread application in protein mutation and design problems in which there are no rich homologous sequences and rapid prediction is needed. In this paper, we constructed two standard datasets for orphan and de novo proteins which have insufficient/none homology information, called Orphan62 and Design204, respectively, to fairly evaluate the performance of the various methods in this case. Then, depending on whether or not utilizing scarce MSA information, we summarized two approaches, MSA-enhanced and MSA-free methods, to effectively solve the issue without sufficient MSAs. MSA-enhanced model aims to improve poor MSA quality from the data source by knowledge distillation and generation models. MSA-free model directly learns the relationship between residues on enormous protein sequences from pre-trained models, bypassing the step of extracting the residue pair representation from MSA. Next, we evaluated the performance of four MSA-free methods (trRosettaX-Single, TRFold, ESMFold and ProtT5) and MSA-enhanced (Bagging MSA) method compared with a traditional MSA-based method AlphaFold2, in two protein structure-related prediction tasks, respectively. Comparison analyses show that trRosettaX-Single and ESMFold which belong to MSA-free method can achieve fast prediction ($\sim\! 40$s) and comparable performance compared with AF2 in tertiary structure prediction, especially for short peptides, $\alpha $-helical segments and targets with few homologous sequences. Bagging MSA utilizing MSA enhancement improves the accuracy of our trained base model which is an MSA-based method when poor homology information exists in secondary structure prediction. Our study provides biologists an insight of how to select rapid and appropriate prediction tools for enzyme engineering and peptide drug development. CONTACT: guofei@csu.edu.cn, jj.tang@siat.ac.cn.

An AlphaFold2 map of the 53BP1 pathway identifies a direct SHLD3-RIF1 interaction critical for shieldin activity.

53BP1 is a chromatin-binding protein that promotes DNA double-strand break repair through the recruitment of downstream effectors including RIF1, shieldin, and CST. The structural basis of the protein-protein interactions within the 53BP1-RIF1-shieldin-CST pathway that are essential for its DNA repair activity is largely unknown. Here, we used AlphaFold2-Multimer (AF2) to predict all possible pairwise combinations of proteins within this pathway and provide structural models of seven previously characterized interactions. This analysis also predicted an entirely novel binding interface between the HEAT-repeat domain of RIF1 and the eIF4E-like domain of SHLD3. Extensive interrogation of this interface through both in vitro pulldown analysis and cellular assays supports the AF2-predicted model and demonstrates that RIF1-SHLD3 binding is essential for shieldin recruitment to sites of DNA damage, and for its role in antibody class switch recombination and PARP inhibitor sensitivity. Direct physical interaction between RIF1 and SHLD3 is therefore essential for 53BP1-RIF1-shieldin-CST pathway activity.

AlphaFold2: A versatile tool to predict the appearance of functional adaptations in evolution: Profilin interactions in uncultured Asgard archaea: Profilin interactions in uncultured Asgard archaea.

The release of AlphaFold2 (AF2), a deep-learning-aided, open-source protein structure prediction program, from DeepMind, opened a new era of molecular biology. The astonishing improvement in the accuracy of the structure predictions provides the opportunity to characterize protein systems from uncultured Asgard archaea, key organisms in evolutionary biology. Despite the accumulation in metagenomics-derived Asgard archaea eukaryotic-like protein sequences, limited structural and biochemical information have restricted the insight in their potential functions. In this review, we focus on profilin, an actin-dynamics regulating protein, which in eukaryotes, modulates actin polymerization through (1) direct actin interaction, (2) polyproline binding, and (3) phospholipid binding. We assess AF2-predicted profilin structures in their potential abilities to participate in these activities. We demonstrate that AF2 is a powerful new tool for understanding the emergence of biological functional traits in evolution.

The AF-2 cofactor binding region is key for the selective SUMOylation of estrogen receptor alpha by antiestrogens.

Antiestrogens (AEs) are used to treat all stages of estrogen receptor (ER)-positive breast cancer. Selective estrogen receptor modulators such as tamoxifen have tissue-specific partial agonist activity, while selective estrogen receptor downregulators such as fulvestrant (ICI182,780) display a more complete antiestrogenic profile. We have previously observed that fulvestrant-induced ERα SUMOylation contributes to transcriptional suppression, but whether this effect is seen with other AEs and is specific to ERα is unclear. Here we show that several AEs induce SUMOylation of ERα, but not ERß, at different levels. Swapping domains between ERα and ERß indicates that the ERα identity of the ligand-binding domain helices 3 and 4 (H3-H4 region), which contribute to the static part of the activation function-2 (AF-2) cofactor binding groove, is sufficient to confer fulvestrant-induced SUMOylation to ERß. This region does not contain lysine residues unique to ERα, suggesting that ERα-specific residues in H3-H4 determine the capacity of the AE-bound ERα ligand-binding domain to recruit the SUMOylation machinery. We also show that the SUMO E3 ligase protein inhibitor of activated STAT 1 increases SUMOylation of ERα and of ERß containing the H3-H4 region of ERα, but not of ERß. Together, these results shed new light on the molecular basis for the differential capacity of selective estrogen receptor modulators and selective estrogen receptor downregulators to suppress transcription by ERα.

Outer membrane ß-barrel structure prediction through the lens of AlphaFold2.

Most proteins found in the outer membrane of gram-negative bacteria share a common domain: the transmembrane ß-barrel. These outer membrane ß-barrels (OMBBs) occur in multiple sizes and different families with a wide range of functions evolved independently by amplification from a pool of homologous ancestral ßß-hairpins. This is part of the reason why predicting their three-dimensional (3D) structure, especially by homology modeling, is a major challenge. Recently, DeepMind's AlphaFold v2 (AF2) became the first structure prediction method to reach close-to-experimental atomic accuracy in CASP even for difficult targets. However, membrane proteins, especially OMBBs, were not abundant during their training, raising the question of how accurate the predictions are for these families. In this study, we assessed the performance of AF2 in the prediction of OMBBs and OMBB-like folds of various topologies using an in-house-developed tool for the analysis of OMBB 3D structures, and barrOs. In agreement with previous studies on other membrane protein classes, our results indicate that AF2 predicts transmembrane ß-barrel structures at high accuracy independently of the use of templates, even for novel topologies absent from the training set. These results provide confidence on the models generated by AF2 and open the door to the structural elucidation of novel transmembrane ß-barrel topologies identified in high-throughput OMBB annotation studies or designed de novo.

Co-evolution at protein-protein interfaces guides inference of stoichiometry of oligomeric protein complexes by de novo structure prediction.

The quaternary structure with specific stoichiometry is pivotal to the specific function of protein complexes. However, determining the structure of many protein complexes experimentally remains a major bottleneck. Structural bioinformatics approaches, such as the deep learning algorithm Alphafold2-multimer (AF2-multimer), leverage the co-evolution of amino acids and sequence-structure relationships for accurate de novo structure and contact prediction. Pseudo-likelihood maximization direct coupling analysis (plmDCA) has been used to detect co-evolving residue pairs by statistical modeling. Here, we provide evidence that combining both methods can be used for de novo prediction of the quaternary structure and stoichiometry of a protein complex. We achieve this by augmenting the existing AF2-multimer confidence metrics with an interpretable score to identify the complex with an optimal fraction of native contacts of co-evolving residue pairs at intermolecular interfaces. We use this strategy to predict the quaternary structure and non-trivial stoichiometries of Bacillus subtilis spore germination protein complexes with unknown structures. Co-evolution at intermolecular interfaces may therefore synergize with AI-based de novo quaternary structure prediction of structurally uncharacterized bacterial protein complexes.

Exploring Host-Binding Machineries of Mycobacteriophages with AlphaFold2.

Although more than 12,000 bacteriophages infecting mycobacteria (mycobacteriophages) have been isolated so far, there is a knowledge gap on their structure-function relationships. Here, we have explored the architecture of host-binding machineries from seven representative mycobacteriophages of the Siphoviridae family infecting Mycobacterium smegmatis, Mycobacterium abscessus, and Mycobacterium tuberculosis, using AlphaFold2 (AF2). AF2 enables confident structural analyses of large and flexible biological assemblies resistant to experimental methods, thereby opening new avenues to shed light on phage structure and function. Our results highlight the modularity and structural diversity of siphophage host-binding machineries that recognize host-specific receptors at the onset of viral infection. Interestingly, the studied mycobacteriophages' host-binding machineries present unique features compared with those of phages infecting other Gram-positive actinobacteria. Although they all assemble the classical Dit (distal tail), Tal (tail-associated lysin), and receptor-binding proteins, five of them contain two potential additional adhesion proteins. Moreover, we have identified brush-like domains formed of multiple polyglycine helices which expose hydrophobic residues as potential receptor-binding domains. These polyglycine-rich domains, which have been observed in only five native proteins, may be a hallmark of mycobacteriophages' host-binding machineries, and they may be more common in nature than expected. Altogether, the unique composition of mycobacteriophages' host-binding machineries indicate they might have evolved to bind to the peculiar mycobacterial cell envelope, which is rich in polysaccharides and mycolic acids. This work provides a rational framework to efficiently produce recombinant proteins or protein domains and test their host-binding function and, hence, to shed light on molecular mechanisms used by mycobacteriophages to infect their host. IMPORTANCE Mycobacteria include both saprophytes, such as the model system Mycobacterium smegmatis, and pathogens, such as Mycobacterium tuberculosis and Mycobacterium abscessus, that are poorly responsive to antibiotic treatments and pose a global public health problem. Mycobacteriophages have been collected at a very large scale over the last decade, and they have proven to be valuable tools for mycobacteria genetic manipulation, rapid diagnostics, and infection treatment. Yet, molecular mechanisms used by mycobacteriophages to infect their host remain poorly understood. Therefore, exploring the structural diversity of mycobacteriophages' host-binding machineries is important not only to better understand viral diversity and bacteriophage-host interactions, but also to rationally develop biotechnological tools. With the powerful protein structure prediction software AlphaFold2, which was publicly released a year ago, it is now possible to gain structural and functional insights on such challenging assemblies.

Probiotic potential and wound-healing activity of Pediococcus pentosaceus strain AF2 isolated from Herniaria glabra L. which is traditionally used to make yogurt.

Probiotics have been a part of our lives for centuries, primarily through fermented foods. They find applications in various fields such as food, healthcare, and agriculture. Nowadays, their utilization is expanding, highlighting the importance of discovering new bacterial strains with probiotic properties suitable for diverse applications. In this study, our aim was to isolate new probiotic bacteria. Herniaria glabra L., a plant traditionally used for yogurt making in some regions and recognized in official medicine in many countries, was chosen as the source for obtaining probiotic bacteria. We conducted bacterial isolation from the plant, molecularly identified the isolated bacteria using 16S rRNA sequencing, characterized their probiotic properties, and assessed their wound-healing effects. As a result of these studies, we identified the bacterium isolated from the plant as Pediococcus pentosaceus strain AF2. We found that the strain AF2 exhibited high resistance to conditions within the gastrointestinal tract. Our reliability analysis showed that the isolate had γ-hemolytic activity and displayed sensitivity to certain tested antibiotics. At the same time, AF2 did not show gelatinase and DNase activity. We observed that the strain AF2 produced metabolites with inhibitory activity against E. coli, B. subtilis, P. vulgaris, S. typhimurium, P. aeruginosa, K. pneumoniae, E. cloacae, and Y. pseudotuberculosis. The auto-aggregation value of the strain AF2 was calculated at 73.44%. Coaggregation values against E. coli and L. monocytogenes bacteria were determined to be 56.8% and 57.38%, respectively. Finally, we tested the wound-healing effect of the strain AF2 with cell culture studies and found that the strain AF2 promoted wound healing.

Improving furfural tolerance in a xylose-fermenting yeast Spathaspora passalidarum CMUWF1-2 via adaptive laboratory evolution.

BACKGROUND: Spathaspora passalidarum is a yeast with the highly effective capability of fermenting several monosaccharides in lignocellulosic hydrolysates, especially xylose. However, this yeast was shown to be sensitive to furfural released during pretreatment and hydrolysis processes of lignocellulose biomass. We aimed to improve furfural tolerance in a previously isolated S. passalidarum CMUWF1-2, which presented thermotolerance and no detectable glucose repression, via adaptive laboratory evolution (ALE). RESULTS: An adapted strain, AF2.5, was obtained from 17 sequential transfers of CMUWF1-2 in YPD broth with gradually increasing furfural concentration. Strain AF2.5 could tolerate higher concentrations of furfural, ethanol and 5-hydroxymethyl furfuraldehyde (HMF) compared with CMUWF1-2 while maintaining the ability to utilize glucose and other sugars simultaneously. Notably, the lag phase of AF2.5 was 2 times shorter than that of CMUWF1-2 in the presence of 2.0 g/l furfural, which allowed the highest ethanol titers to be reached in a shorter period. To investigate more in-depth effects of furfural, intracellular reactive oxygen species (ROS) accumulation was observed and, in the presence of 2.0 g/l furfural, AF2.5 exhibited 3.41 times less ROS accumulation than CMUWF1-2 consistent with the result from nuclear chromatins diffusion, which the cells number of AF2.5 with diffuse chromatins was also 1.41 and 1.24 times less than CMUWF1-2 at 24 and 36 h, respectively. CONCLUSIONS: An enhanced furfural tolerant strain of S. passalidarum was achieved via ALE techniques, which shows faster and higher ethanol productivity than that of the wild type. Not only furfural tolerance but also ethanol and HMF tolerances were improved.

Conservation of Hot Spots and Ligand Binding Sites in Protein Models by AlphaFold2.

The neural network-based program AlphaFold2 (AF2) provides high accuracy structure prediction for a large fraction of globular proteins. An important question is whether these models are accurate enough for reliably docking small ligands. Several recent papers and the results of CASP15 reveal that local conformational errors reduce the success rates of direct ligand docking. Here, we focus on the ability of the models to conserve the location of binding hot spots, regions on the protein surface that significantly contribute to the binding free energy of the protein-ligand interaction. Clusters of hot spots predict the location and even the druggability of binding sites, and hence are important for computational drug discovery. The hot spots are determined by protein mapping that is based on the distribution of small fragment-sized probes on the protein surface and is less sensitive to local conformation than docking. Mapping models taken from the AlphaFold Protein Structure Database show that identifying binding sites is more reliable than docking, but the success rates are still 5% to 10% lower than based on mapping X-ray structures. The drop in accuracy is particularly large for models of multidomain proteins. However, both the model binding sites and the mapping results can be substantially improved by generating AF2 models for the ligand binding domains of interest rather than the entire proteins and even more if using forced sampling with multiple initial seeds. The mapping of such models tends to reach the accuracy of results obtained by mapping the X-ray structures.

A Perspective on the Prospective Use of AI in Protein Structure Prediction.

AlphaFold2 (AF2) and RoseTTaFold (RF) have revolutionized structural biology, serving as highly reliable and effective methods for predicting protein structures. This article explores their impact and limitations, focusing on their integration into experimental pipelines and their application in diverse protein classes, including membrane proteins, intrinsically disordered proteins (IDPs), and oligomers. In experimental pipelines, AF2 models help X-ray crystallography in resolving the phase problem, while complementarity with mass spectrometry and NMR data enhances structure determination and protein flexibility prediction. Predicting the structure of membrane proteins remains challenging for both AF2 and RF due to difficulties in capturing conformational ensembles and interactions with the membrane. Improvements in incorporating membrane-specific features and predicting the structural effect of mutations are crucial. For intrinsically disordered proteins, AF2's confidence score (pLDDT) serves as a competitive disorder predictor, but integrative approaches including molecular dynamics (MD) simulations or hydrophobic cluster analyses are advocated for accurate dynamics representation. AF2 and RF show promising results for oligomeric models, outperforming traditional docking methods, with AlphaFold-Multimer showing improved performance. However, some caveats remain in particular for membrane proteins. Real-life examples demonstrate AF2's predictive capabilities in unknown protein structures, but models should be evaluated for their agreement with experimental data. Furthermore, AF2 models can be used complementarily with MD simulations. In this Perspective, we propose a "wish list" for improving deep-learning-based protein folding prediction models, including using experimental data as constraints and modifying models with binding partners or post-translational modifications. Additionally, a meta-tool for ranking and suggesting composite models is suggested, driving future advancements in this rapidly evolving field.

Template-free prediction of a new monotopic membrane protein fold and assembly by AlphaFold2.

AlphaFold2 (AF2) has revolutionized the field of protein structural prediction. Here, we test its ability to predict the tertiary and quaternary structure of a previously undescribed scaffold with new folds and unusual architecture, the monotopic membrane protein caveolin-1 (CAV1). CAV1 assembles into a disc-shaped oligomer composed of 11 symmetrically arranged protomers, each assuming an identical new fold, and contains the largest parallel ß-barrel known to exist in nature. Remarkably, AF2 predicts both the fold of the protomers and the interfaces between them. It also assembles between seven and 15 copies of CAV1 into disc-shaped complexes. However, the predicted multimers are energetically strained, especially the parallel ß-barrel. These findings highlight the ability of AF2 to correctly predict new protein folds and oligomeric assemblies at a granular level while missing some elements of higher-order complexes, thus positing a new direction for the continued development of deep-learning protein structure prediction approaches.

Structures of Plasmodium falciparum Chloroquine Resistance Transporter (PfCRT) Isoforms and Their Interactions with Chloroquine.

Using a recently elucidated atomic-resolution cryogenic electron microscopy (cryo-EM) structure for the Plasmodium falciparum chloroquine resistance transporter (PfCRT) protein 7G8 isoform as template [Kim, J.; Nature 2019, 576, 315-320], we use Monte Carlo molecular dynamics (MC/MD) simulations of PfCRT embedded in a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) membrane to solve energy-minimized structures for 7G8 PfCRT and two additional PfCRT isoforms that harbor 5 or 7 amino acid substitutions relative to 7G8 PfCRT. Guided by drug binding previously defined using chloroquine (CQ) photoaffinity probe labeling, we also use MC/MD energy minimization to elucidate likely CQ binding geometries for the three membrane-embedded isoforms. We inventory salt bridges and hydrogen bonds in these structures and summarize how the limited changes in primary sequence subtly perturb local PfCRT isoform structure. In addition, we use the "AlphaFold" artificial intelligence AlphaFold2 (AF2) algorithm to solve for domain structure that was not resolved in the previously reported 7G8 PfCRT cryo-EM structure, and perform MC/MD energy minimization for the membrane-embedded AF2 structures of all three PfCRT isoforms. We compare energy-minimized structures generated using cryo-EM vs AF2 templates. The results suggest how amino acid substitutions in drug resistance-associated isoforms of PfCRT influence PfCRT structure and CQ transport.

Improved multimer prediction using massive sampling with AlphaFold in CASP15.

AlphaFold2 has revolutionized structure prediction by achieving high accuracy comparable to experimentally determined structures. However, there is still room for improvement, especially for challenging cases like multimers. A key to the success of AlphaFold is its ability to assess and rank its own predictions. Our basic idea for the Wallner group in CASP15 was to exploit this excellent scoring function in AlphaFold by massive sampling. To achieve this goal, we conducted AlphaFold runs using six different settings, using templates, without templates, and with an increased number of recycles for both multimer v1 and v2 weights. In all instances, we enabled dropout layers during inference, allowing for sampling of uncertainty and enhancing the diversity of the generated models. In total, 274 289 models were generated for the 38 targets in CASP15, with a median of 4810 models per target. Of these 38 targets, 10 were high quality, 11 were medium quality, 11 were acceptable, and only 6 were incorrect. The improvement over the baseline method, NBIS-AF2-multimer, is substantial, with the mean DockQ increasing from 0.43 to 0.56, with several targets showing a DockQ score increase of +0.6 units. Remarkable, considering Wallner and NBIS-AF2-multimer were using identical input data. The success can be attributed to the diversified sampling using dropout with different settings and, in particular, the use of multimer v1, which is much more susceptible to sampling compared with v2. The method is available here: http://wallnerlab.org/AFsample/.

Improved protein structure prediction with trRosettaX2, AlphaFold2, and optimized MSAs in CASP15.

We present the monomer and multimer structure prediction results of our methods in CASP15. We first designed an elaborate pipeline that leverages complementary sequence databases and advanced database searching algorithms to generate high-quality multiple sequence alignments (MSAs). Top MSAs were then selected for the subsequent step of structure prediction. We utilized trRosettaX2 and AlphaFold2 for monomer structure prediction (group name Yang-Server), and AlphaFold-Multimer for multimer structure prediction (group name Yang-Multimer). Yang-Server and Yang-Multimer are ranked at the top and the fourth, respectively, for monomer and multimer structure prediction. For 94 monomers, the average TM-score of the predicted structure models by Yang-Server is 0.876, compared to 0.798 by the default AlphaFold2 (i.e., the group NBIS-AF2-standard). For 42 multimers, the average DockQ score of the predicted structure models by Yang-Multimer is 0.464, compared to 0.389 by the default AlphaFold-Multimer (i.e., the group NBIS-AF2-multimer). Detailed analysis of the results shows that several factors contribute to the improvement, including improved MSAs, iterated modeling for large targets, interplay between monomer and multimer structure prediction for intertwined structures, etc. However, the structure predictions for orphan proteins and multimers remain challenging, and breakthroughs in this area are anticipated in the future.

Tertiary structure assessment at CASP15.

The results of tertiary structure assessment at CASP15 are reported. For the first time, recognizing the outstanding performance of AlphaFold 2 (AF2) at CASP14, all single-chain predictions were assessed together, irrespective of whether a template was available. At CASP15, there was no single stand-out group, with most of the best-scoring groups-led by PEZYFoldings, UM-TBM, and Yang Server-employing AF2 in one way or another. Many top groups paid special attention to generating deep Multiple Sequence Alignments (MSAs) and testing variant MSAs, thereby allowing them to successfully address some of the hardest targets. Such difficult targets, as well as lacking templates, were typically proteins with few homologues. Local divergence between prediction and target correlated with localization at crystal lattice or chain interfaces, and with regions exhibiting high B-factor factors in crystal structure targets, and should not necessarily be considered as representing error in the prediction. However, analysis of exposed and buried side chain accuracy showed room for improvement even in the latter. Nevertheless, a majority of groups produced high-quality predictions for most targets, which are valuable for experimental structure determination, functional analysis, and many other tasks across biology. These include those applying methods similar to those used to generate major resources such as the AlphaFold Protein Structure Database and the ESM Metagenomic atlas: the confidence estimates of the former were also notably accurate.

The impact of AI-based modeling on the accuracy of protein assembly prediction: Insights from CASP15.

In CASP15, 87 predictors submitted around 11 000 models on 41 assembly targets. The community demonstrated exceptional performance in overall fold and interface contact predictions, achieving an impressive success rate of 90% (compared to 31% in CASP14). This remarkable accomplishment is largely due to the incorporation of DeepMind's AF2-Multimer approach into custom-built prediction pipelines. To evaluate the added value of participating methods, we compared the community models to the baseline AF2-Multimer predictor. In over 1/3 of cases, the community models were superior to the baseline predictor. The main reasons for this improved performance were the use of custom-built multiple sequence alignments, optimized AF2-Multimer sampling, and the manual assembly of AF2-Multimer-built subcomplexes. The best three groups, in order, are Zheng, Venclovas, and Wallner. Zheng and Venclovas reached a 73.2% success rate over all (41) cases, while Wallner attained 69.4% success rate over 36 cases. Nonetheless, challenges remain in predicting structures with weak evolutionary signals, such as nanobody-antigen, antibody-antigen, and viral complexes. Expectedly, modeling large complexes also remains challenging due to their high memory compute demands. In addition to the assembly category, we assessed the accuracy of modeling interdomain interfaces in the tertiary structure prediction targets. Models on seven targets featuring 17 unique interfaces were analyzed. Best predictors achieved a 76.5% success rate, with the UM-TBM group being the leader. In the interdomain category, we observed that the predictors faced challenges, as in the case of the assembly category, when the evolutionary signal for a given domain pair was weak or the structure was large. Overall, CASP15 witnessed unprecedented improvement in interface modeling, reflecting the AI revolution seen in CASP14.