Protein Structure Prediction Methods Research Articles

AlphaFold2 Predicts Alternative Conformation Populations in Green Fluorescent Protein Variants.

Artificial intelligence-based protein structure prediction methods such as AlphaFold2 have emerged as powerful tools for characterizing proteins sequence-structure relationship offering unprecedented opportunities for the molecular interpretation of biological and biochemical phenomena. While initially confined to providing a static representation of proteins through their global free-energy minimum, AlphaFold2 has demonstrated the ability to partially sample conformational landscapes, providing insights into protein dynamics, which is fundamental for interpreting and potentially tuning the function of natural and artificial proteins. In this study, we show that targeted column masking of AlphaFold2's multiple sequence alignment enables the characterization and estimation of the population ratio of the two main conformations of engineered green fluorescent proteins with alternative β-strands. The possibility of quickly estimating relative populations through AlphaFold2 predictions is expected to speed-up the computational design of related systems for sensing applications.

Deep learning methods for protein structure prediction

AbstractProtein structure prediction (PSP) has been a prominent topic in bioinformatics and computational biology, aiming to predict protein function and structure from sequence data. The three‐dimensional conformation of proteins is pivotal for their intricate biological roles. With the advancement of computational capabilities and the adoption of deep learning (DL) technologies (especially Transformer network architectures), the PSP field has ushered in a brand‐new era of “neuralization.” Here, we focus on reviewing the evolution of PSP from traditional to modern deep learning‐based approaches and the characteristics of various structural prediction methods. This emphasizes the advantages of deep learning‐based hybrid prediction methods over traditional approaches. This study also provides a summary analysis of widely used bioinformatics databases and the latest structure prediction models. It discusses deep learning networks and algorithmic optimization for model training, validation, and evaluation. In addition, a summary discussion of the major advances in deep learning‐based protein structure prediction is presented. The update of AlphaFold 3 further extends the boundaries of prediction models, especially in protein‐small molecule structure prediction. This marks a key shift toward a holistic approach in biomolecular structure elucidation, aiming at solving almost all sequence‐to‐structure puzzles in various biological phenomena.

A comprehensive exploration of the druggable conformational space of protein kinases using AI-predicted structures.

Protein kinase function and interactions with drugs are controlled in part by the movement of the DFG and ɑC-Helix motifs that are related to the catalytic activity of the kinase. Small molecule ligands elicit therapeutic effects with distinct selectivity profiles and residence times that often depend on the active or inactive kinase conformation(s) they bind. Modern AI-based structural modeling methods have the potential to expand upon the limited availability of experimentally determined kinase structures in inactive states. Here, we first explored the conformational space of kinases in the PDB and models generated by AlphaFold2 (AF2) and ESMFold, two prominent AI-based protein structure prediction methods. Our investigation of AF2's ability to explore the conformational diversity of the kinome at various multiple sequence alignment (MSA) depths showed a bias within the predicted structures of kinases in DFG-in conformations, particularly those controlled by the DFG motif, based on their overabundance in the PDB. We demonstrate that predicting kinase structures using AF2 at lower MSA depths explored these alternative conformations more extensively, including identifying previously unobserved conformations for 398 kinases. Ligand enrichment analyses for 23 kinases showed that, on average, docked models distinguished between active molecules and decoys better than random (average AUC (avgAUC) of 64.58), but select models perform well (e.g., avgAUCs for PTK2 and JAK2 were 79.28 and 80.16, respectively). Further analysis explained the ligand enrichment discrepancy between low- and high-performing kinase models as binding site occlusions that would preclude docking. The overall results of our analyses suggested that, although AF2 explored previously uncharted regions of the kinase conformational space and select models exhibited enrichment scores suitable for rational drug discovery, rigorous refinement of AF2 models is likely still necessary for drug discovery campaigns.

DeepPPAPredMut: deep ensemble method for predicting the binding affinity change in protein-protein complexes upon mutation.

Protein-protein interactions underpin many cellular processes and their disruption due to mutations can lead to diseases. With the evolution of protein structure prediction methods like AlphaFold2 and the availability of extensive experimental affinity data, there is a pressing need for updated computational tools that can efficiently predict changes in binding affinity caused by mutations in protein-protein complexes. We developed a deep ensemble model that leverages protein sequences, predicted structure-based features, and protein functional classes to accurately predict the change in binding affinity due to mutations. The model achieved a correlation of 0.97 and a mean absolute error (MAE) of 0.35 kcal/mol on the training dataset, and maintained robust performance on the test set with a correlation of 0.72 and a MAE of 0.83 kcal/mol. Further validation using Leave-One-Out Complex (LOOC) cross-validation exhibited a correlation of 0.83 and a MAE of 0.51 kcal/mol, indicating consistent performance. https://web.iitm.ac.in/bioinfo2/DeepPPAPredMut/index.html.

Investigation of the MDM2-binding potential of de novo designed peptides using enhanced sampling simulations

The tumor suppressor p53 plays a crucial role in cellular responses to various stresses, regulating key processes such as apoptosis, senescence, and DNA repair. Dysfunctional p53, prevalent in approximately 50 % of human cancers, contributes to tumor development and resistance to treatment. This study employed deep learning-based protein design and structure prediction methods to identify novel high-affinity peptide binders (Pep1 and Pep2) targeting MDM2, with the aim of disrupting its interaction with p53. Extensive all-atom molecular dynamics simulations highlighted the stability of the designed peptide in complex with the target, supported by several structural analyses, including RMSD, RMSF, Rg, SASA, PCA, and free energy landscapes. Using the steered molecular dynamics and umbrella sampling simulations, we elucidate the dissociation dynamics of p53, Pep1, and Pep2 from MDM2. Notable differences in interaction profiles were observed, emphasizing the distinct dissociation patterns of each peptide. In conclusion, the results of our umbrella sampling simulations suggest Pep1 as a higher-affinity MDM2 binder compared to p53 and Pep2, positioning it as a potential inhibitor of the MDM2-p53 interaction. Using state-of-the-art protein design tools and advanced MD simulations, this study provides a comprehensive framework for rational in silico design of peptide binders with therapeutic implications in disrupting MDM2-p53 interactions for anticancer interventions.

Fast, accurate ranking of engineered proteins by target-binding propensity using structure modeling

Deep learning-based methods for protein structure prediction have achieved unprecedented accuracy. Yet, their utility in the engineering of protein-based binders remains constrained due to a gap between the ability to predict the structures of candidate proteins and the ability to assess which of those proteins are more probable to bind to a target. To bridge this gap, we introduce Automated Pairwise Peptide-Receptor AnalysIs for Screening Engineered proteins (APPRAISE), a method for predicting the target binding propensity of engineered proteins. After generating models of engineered proteins competing for binding to a target using an established structure-prediction tool such as AlphaFold-Multimer or ESMFold, APPRAISE performs a rapid (under 1 CPU second per model) scoring analysis that takes into account biophysical and geometrical constraints. As proof-of-concept cases, we demonstrate that APPRAISE can accurately classify receptor-dependent vs. receptor-independent adeno-associated viral vectors and diverse classes of engineered proteins such as miniproteins targeting the SARS-CoV-2 spike, nanobodies targeting a G-protein-coupled receptor, and peptides that specifically bind to transferrin receptor or PD-L1. APPRAISE is accessible through a web-based notebook interface using Google Colaboratory (https://tiny.cc/APPRAISE). With its accuracy, interpretability, and generalizability, APPRAISE promises to expand the utility of protein structure prediction and accelerate protein engineering for biomedical applications.

Can Protein Structure Prediction Methods Capture Alternative Conformations of Membrane Transporters?

Understanding the conformational dynamics of proteins, such as the inward-facing (IF) and outward-facing (OF) transition observed in transporters, is vital for elucidating their functional mechanisms. Despite significant advances in protein structure prediction (PSP) over the past three decades, most efforts have been focused on single-state prediction, leaving multistate or alternative conformation prediction (ACP) relatively unexplored. This discrepancy has led to the development of highly accurate PSP methods such as AlphaFold, yet their capabilities for ACP remain limited. To investigate the performance of current PSP methods in ACP, we curated a data set, named IOMemP, consisting of 32 experimentally determined high-resolution IF and OF structures of 16 membrane proteins with substantial conformational changes. We benchmarked 12 representative PSP methods, along with two recent multistate methods based on AlphaFold, against this data set. Our findings reveal a remarkably consistent preference for specific states across various PSP methods. We elucidated how coevolution information in MSAs influences state preference. Moreover, we showed that AlphaFold, when excluding coevolution information, estimated similar energies between the experimental IF and OF conformations, indicating that the energy model learned by AlphaFold is not biased toward any particular state. Our IOMemP data set and benchmark results are anticipated to advance the development of robust ACP methods.

Enhancing cryo-EM structure prediction with DeepTracer and AlphaFold2 integration.

Understanding the protein structures is invaluable in various biomedical applications, such as vaccine development. Protein structure model building from experimental electron density maps is a time-consuming and labor-intensive task. To address the challenge, machine learning approaches have been proposed to automate this process. Currently, the majority of the experimental maps in the database lack atomic resolution features, making it challenging for machine learning-based methods to precisely determine protein structures from cryogenic electron microscopy density maps. On the other hand, protein structure prediction methods, such as AlphaFold2, leverage evolutionary information from protein sequences and have recently achieved groundbreaking accuracy. However, these methods often require manual refinement, which is labor intensive and time consuming. In this study, we present DeepTracer-Refine, an automated method that refines AlphaFold predicted structures by aligning them to DeepTracers modeled structure. Our method was evaluated on 39 multi-domain proteins and we improved the average residue coverage from 78.2 to 90.0% and average local Distance Difference Test score from 0.67 to 0.71. We also compared DeepTracer-Refine with Phenixs AlphaFold refinement and demonstrated that our method not only performs better when the initial AlphaFold model is less precise but also surpasses Phenix in run-time performance.

Abstract 1252: AI-derived predictions improve identification of real-world cancer driver mutations

Abstract AI-based variant effect predictors (VEPs), such as AlphaMissense, which is based on the protein structure prediction method AlphaFold, have gained significant attention for their potential to predict pathogenic effects of mutations. However, the utility of these methods for identifying pathogenic mutations in cancer remains unknown. We sought to quantify the prevalence of variants of unknown significance (VUSs) across the pan-cancer, multi-institution AACR GENIE cohort (N=160,965 samples) and test the utility of VEPs (SIFT, PolyPhen, MutationAssessor, REVEL, CADD, and AlphaMissense), using publicly available mutation annotations from each method, for 1) Annotating known pathogenic somatic cancer variants across the GENIE cohort, using known benign single nucleotide polymorphisms as negative controls 2) Identifying VUSs associated with overall survival (OS) in two cohorts of patients with non-small cell cancer (NSCLC), the MSK-IMPACT NCSLC (N=8,690) and non-MSK AACR GENIE Biopharma Collaborative NSCLC (N=977) cohorts, and 3) Identifying VUSs associated with known genomic patterns of mutual exclusivity in the MSK-IMPACT NSCLC cohort. In the GENIE cohort, 79% of mutations identified were VUSs according to an FDA-recognized molecular knowledge database, with a wide range of VUS frequencies across genes. Among VEPs trained without prior human-annotated knowledge, CADD and AlphaMissense had the highest AUROCs (0.940, 95%CI 0.937-0.943 and 0.917, 95%CI 0.913-0.922 respectively) for predicting known cancer drivers, demonstrating the power of protein structure modeling or functional genomic data for enhancing cancer drivers prediction. All VEPs identified known driver mutations in tumor suppressor genes more accurately than oncogenes, highlighting the challenge in identifying gain-of-function mutations. We tested whether VUSs reclassified as pathogenic by VEPs are associated with prognosis using inverse probability of treatment-weighted Cox’s propensity hazard models controlling for tumor mutational burden, treatment history, and other factors. VUSs reclassified as pathogenic by AlphaMissense and other VEPs in KEAP1 and SMARCA4 correlated with worse OS, comparable to that of known oncogenic mutations in those genes. Patients with tumors containing VUSs annotated as benign had similar OS to patients without analogous driver mutations. Finally, VUSs reclassified as oncogenic by AlphaMissense in genes within the RTK/RAS and NRF2 pathways followed expected patterns of mutual exclusivity, further suggesting biological validity. Despite not being trained to predict somatic effects in cancer, AI-derived mutation annotations can broaden the subset of annotated pathogenic variants in cancer and contribute to a more complete understanding of cancer genetics. Real-world outcomes and alteration patterns are additional benchmarks to consider when assessing the utility of VEPs in cancer. Citation Format: Thinh N. Tran, Chris Fong, Karl Pichotta, Anisha Luthra, Ronglai Shen, Yuan Chen, Michele Waters, Susie Kim, Gregory Riely, Debyani Chakravarty, Nikolaus Schultz, Justin Jee. AI-derived predictions improve identification of real-world cancer driver mutations [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 1252.

Single-sequence protein structure prediction by integrating protein language models

Protein structure prediction has been greatly improved by deep learning in the past few years. However, the most successful methods rely on multiple sequence alignment (MSA) of the sequence homologs of the protein under prediction. In nature, a protein folds in the absence of its sequence homologs and thus, a MSA-free structure prediction method is desired. Here, we develop a single-sequence-based protein structure prediction method RaptorX-Single by integrating several protein language models and a structure generation module and then study its advantage over MSA-based methods. Our experimental results indicate that in addition to running much faster than MSA-based methods such as AlphaFold2, RaptorX-Single outperforms AlphaFold2 and other MSA-free methods in predicting the structure of antibodies (after fine-tuning on antibody data), proteins of very few sequence homologs, and single mutation effects. By comparing different protein language models, our results show that not only the scale but also the training data of protein language models will impact the performance. RaptorX-Single also compares favorably to MSA-based AlphaFold2 when the protein under prediction has a large number of sequence homologs.

The structural landscape of the immunoglobulin fold by large-scale de novo design.

De novo designing immunoglobulin-like frameworks that allow for functional loop diversification shows great potential for crafting antibody-like scaffolds with fully customizable structures and functions. In this work, we combined de novo parametric design with deep-learning methods for protein structure prediction and design to explore the structural landscape of 7-stranded immunoglobulin domains. After screening folding of nearly 4 million designs, we have assembled a structurally diverse library of ~50,000 immunoglobulin domains with high-confidence AlphaFold2 predictions and structures diverging from naturally occurring ones. The designed dataset enabled us to identify structural requirements for the correct folding of immunoglobulin domains, shed light on β-sheet-β-sheet rotational preferences and how these are linked to functional properties. Our approach eliminates the need for preset loop conformations and opens the route to large-scale de novo design of immunoglobulin-like frameworks.

Computational Prediction of Structure, Function, and Interaction of Myzus persicae (Green Peach Aphid) Salivary Effector Proteins.

Similar to plant pathogens, phloem-feeding insects such as aphids deliver effector proteins inside their hosts that act to promote host susceptibility and enable feeding and infestation. Despite exciting progress toward identifying and characterizing effector proteins from these insects, their functions remain largely unknown. The recent groundbreaking development in protein structure prediction algorithms, combined with the availability of proteomics and transcriptomic datasets for agriculturally important pests, provides new opportunities to explore the structural and functional diversity of effector repertoires. In this study, we sought to gain insight into the infection strategy used by the Myzus persicae (green peach aphid) by predicting and analyzing the structures of a set of 71 effector candidate proteins. We used two protein structure prediction methods, AlphaFold and OmegaFold, that produced mutually consistent results. We observed a wide continuous spectrum of structures among the effector candidates, from disordered proteins to globular enzymes. We made use of the structural information and state-of-the-art computational methods to predict M. persicae effector protein properties, including function and interaction with host plant proteins. Overall, our investigation provides novel insights into prediction of structure, function, and interaction of M. persicae effector proteins and will guide the necessary experimental characterization to address new hypotheses. [Formula: see text] Copyright © 2024 The Author(s). This is an open access article distributed under the CC BY-NC-ND 4.0 International license.

Investigating the ability of deep learning-based structure prediction to extrapolate and/or enrich the set of antibody CDR canonical forms.

Deep learning models have been shown to accurately predict protein structure from sequence, allowing researchers to explore protein space from the structural viewpoint. In this paper we explore whether "novel" features, such as distinct loop conformations can arise from these predictions despite not being present in the training data. Here we have used ABodyBuilder2, a deep learning antibody structure predictor, to predict the structures of ~1.5M paired antibody sequences. We examined the predicted structures of the canonical CDR loops and found that most of these predictions fall into the already described CDR canonical form structural space. We also found a small number of "new" canonical clusters composed of heterogeneous sequences united by a common sequence motif and loop conformation. Analysis of these novel clusters showed their origins to be either shapes seen in the training data at very low frequency or shapes seen at high frequency but at a shorter sequence length. To evaluate explicitly the ability of ABodyBuilder2 to extrapolate, we retrained several models whilst withholding all antibody structures of a specific CDR loop length or canonical form. These "starved" models showed evidence of generalisation across CDRs of different lengths, but they did not extrapolate to loop conformations which were highly distinct from those present in the training data. However, the models were able to accurately predict a canonical form even if only a very small number of examples of that shape were in the training data. Our results suggest that deep learning protein structure prediction methods are unable to make completely out-of-domain predictions for CDR loops. However, in our analysis we also found that even minimal amounts of data of a structural shape allow the method to recover its original predictive abilities. We have made the ~1.5 M predicted structures used in this study available to download at https://doi.org/10.5281/zenodo.10280181.

Protein structure accuracy estimation using geometry-complete perceptron networks.

Estimating the accuracy of protein structural models is a critical task in protein bioinformatics. The need for robust methods in the estimation of protein model accuracy (EMA) is prevalent in the field of protein structure prediction, where computationally-predicted structures need to be screened rapidly for the reliability of the positions predicted for each of their amino acid residues and their overall quality. Current methods proposed for EMA are either coupled tightly to existing protein structure prediction methods or evaluate protein structures without sufficiently leveraging the rich, geometric information available in such structures to guide accuracy estimation. In this work, we propose a geometric message passing neural network referred to as the geometry-complete perceptron network for protein structure EMA (GCPNet-EMA), where we demonstrate through rigorous computational benchmarks that GCPNet-EMA's accuracy estimations are 47% faster and more than 10% (6%) more correlated with ground-truth measures of per-residue (per-target) structural accuracy compared to baseline state-of-the-art methods for tertiary (multimer) structure EMA including AlphaFold 2. The source code and data for GCPNet-EMA are available on GitHub, and a public web server implementation is freely available.

An AI-generated proteome-scale dataset of predicted protein structures for the ctenophore Mnemiopsis leidyi.

This Dataset Brief describes the computational prediction of protein structures for the ctenophore Mnemiopsis leidyi. Here, we report the proteome-scale generation of 15,333 protein structure predictions using AlphaFold, as well as an updated implementation of publicly available search, manipulation, and visualization tools for these protein structure predictions through the Mnemiopsis Genome Project Portal (https://research.nhgri.nih.gov/mnemiopsis). The utility of these predictions is demonstrated by highlighting comparisons to experimentally determined structures for the light-sensitive protein mnemiopsin 1 and the ionotropic glutamate receptor (iGluR). The application of these novel protein structure prediction methods will serve to further position non-bilaterian species such as Mnemiopsis as powerful model systems for the study of early animal evolution and human health.

A Rare Variant in MDH2 (rs111879470) Is Associated with Predisposition to Recurrent Breast Cancer in an Extended High-Risk Pedigree.

A significant fraction of breast cancer recurs, with lethal outcome, but specific genetic variants responsible have yet to be identified. Five cousin pairs with recurrent breast cancer from pedigrees with a statistical excess of recurrent breast cancer were sequenced to identify rare, shared candidate predisposition variants. The candidates were tested for association with breast cancer risk with UKBiobank data. Additional breast cancer cases were assayed for a subset of candidate variants to test for co-segregation. Three-dimensional protein structure prediction methods were used to investigate how the mutation under consideration is predicted to change structural and electrostatic properties in the mutated protein. One hundred and eighty-one rare candidate predisposition variants were shared in at least one cousin pair from a high-risk pedigree. A rare variant in MDH2 was found to segregate with breast-cancer-affected relatives in one extended pedigree. MDH2 is an estrogen-stimulated gene encoding the protein malate dehydrogenase, which catalyzes the reversible oxidation of malate to oxaloacetate. The molecular simulation results strongly suggest that the mutation changes the NAD+ binding pocket electrostatics of MDH2. This small sequencing study, using a powerful approach based on recurrent breast cancer cases from high-risk pedigrees, identified a set of strong candidate variants for inherited predisposition for breast cancer recurrence, including MDH2, which should be pursued in other resources.

Comprehensive analysis of protein-protein interactions (PPIs) with structure prediction program for breast cancer determination

Breast cancer is a prevalent disease that primarily affects women, with significant implications for public health. Early detection improves survival rates, and technological advancements can enhance early diagnosis. Protein structure prediction methods can provide valuable insights into the structure of proteins involved in breast cancer, including human epidermal growth factor receptor-2 (HER2). HER2 is a known protein receptor that plays a critical role in breast cancer development and is a target for therapy. Predicting the binding affinity between HER2 and potential ligands can help identify novel treatment options. This study aimed to predict the structure of HER2 using I-TASSER and determine potential ligands using empirical graph neural network-based scoring functions. Molecular docking simulations were performed to evaluate the binding conformation and stability of the HER2-ligand complexes. The results showed the potential ligands identified by I-TASSER: AEE, UUU, and Mg2+. Afterward, their binding affinity with HER2 was assessed, yielding AEE as the best binding with the lowest vina score.

AlphaFold predictions are valuable hypotheses and accelerate but do not replace experimental structure determination

Artificial intelligence-based protein structure prediction methods such as AlphaFold have revolutionized structural biology. The accuracies of these predictions vary, however, and they do not take into account ligands, covalent modifications or other environmental factors. Here, we evaluate how well AlphaFold predictions can be expected to describe the structure of a protein by comparing predictions directly with experimental crystallographic maps. In many cases, AlphaFold predictions matched experimental maps remarkably closely. In other cases, even very high-confidence predictions differed from experimental maps on a global scale through distortion and domain orientation, and on a local scale in backbone and side-chain conformation. We suggest considering AlphaFold predictions as exceptionally useful hypotheses. We further suggest that it is important to consider the confidence in prediction when interpreting AlphaFold predictions and to carry out experimental structure determination to verify structural details, particularly those that involve interactions not included in the prediction.

A sequence-based evolutionary distance method for Phylogenetic analysis of highly divergent proteins

Because of the limited effectiveness of prevailing phylogenetic methods when applied to highly divergent protein sequences, the phylogenetic analysis problem remains challenging. Here, we propose a sequence-based evolutionary distance algorithm termed sequence distance (SD), which innovatively incorporates site-to-site correlation within protein sequences into the distance estimation. In protein superfamilies, SD can effectively distinguish evolutionary relationships both within and between protein families, producing phylogenetic trees that closely align with those based on structural information, even with sequence identity less than 20%. SD is highly correlated with the similarity of the protein structure, and can calculate evolutionary distances for thousands of protein pairs within seconds using a single CPU, which is significantly faster than most protein structure prediction methods that demand high computational resources and long run times. The development of SD will significantly advance phylogenetics, providing researchers with a more accurate and reliable tool for exploring evolutionary relationships.

Critical assessment of methods of protein structure prediction (CASP)-Round XV.

Computing protein structure from amino acid sequence information has been a long-standing grand challenge. Critical assessment of structure prediction (CASP) conducts community experiments aimed at advancing solutions to this and related problems. Experiments are conducted every 2 years. The 2020 experiment (CASP14) saw major progress, with the second generation of deep learning methods delivering accuracy comparable with experiment for many single proteins. There is an expectation that these methods will have much wider application in computational structural biology. Here we summarize results from the most recent experiment, CASP15, in 2022, with an emphasis on new deep learning-driven progress. Other papers in this special issue of proteins provide more detailed analysis. For single protein structures, the AlphaFold2 deep learning method is still superior to other approaches, but there are two points of note. First, although AlphaFold2 was the core of all the most successful methods, there was a wide variety of implementation and combination with other methods. Second, using the standard AlphaFold2 protocol and default parameters only produces the highest quality result for about two thirds of the targets, and more extensive sampling is required for the others. The major advance in this CASP is the enormous increase in the accuracy of computed protein complexes, achieved by the use of deep learning methods, although overall these do not fully match the performance for single proteins. Here too, AlphaFold2 based method perform best, and again more extensive sampling than the defaults is often required. Also of note are the encouraging early results on the use of deep learning to compute ensembles of macromolecular structures. Critically for the usability of computed structures, for both single proteins and protein complexes, deep learning derived estimates of both local and global accuracy are of high quality, however the estimates in interface regions are slightly less reliable. CASP15 also included computation of RNA structures for the first time. Here, the classical approaches produced better agreement with experiment than the new deep learning ones, and accuracy is limited. Also, for the first time, CASP included the computation of protein-ligand complexes, an area of special interest for drug design. Here too, classical methods were still superior to deep learning ones. Many new approaches were discussed at the CASP conference, and it is clear methods will continue to advance.