Structure Prediction Algorithms Research Articles

Exploring the fragmentation efficiency of proteins analyzed by MALDI-TOF-TOF tandem mass spectrometry using computational and statistical analyses.

Matrix-assisted laser desorption/ionization time-of-flight-time-of-flight (MALDI-TOF-TOF) tandem mass spectrometry (MS/MS) is a rapid technique for identifying intact proteins from unfractionated mixtures by top-down proteomic analysis. MS/MS allows isolation of specific intact protein ions prior to fragmentation, allowing fragment ion attribution to a specific precursor ion. However, the fragmentation efficiency of mature, intact protein ions by MS/MS post-source decay (PSD) varies widely, and the biochemical and structural factors of the protein that contribute to it are poorly understood. With the advent of protein structure prediction algorithms such as Alphafold2, we have wider access to protein structures for which no crystal structure exists. In this work, we use a statistical approach to explore the properties of bacterial proteins that can affect their gas phase dissociation via PSD. We extract various protein properties from Alphafold2 predictions and analyze their effect on fragmentation efficiency. Our results show that the fragmentation efficiency from cleavage of the polypeptide backbone on the C-terminal side of glutamic acid (E) and asparagine (N) residues were nearly equal. In addition, we found that the rearrangement and cleavage on the C-terminal side of aspartic acid (D) residues that result from the aspartic acid effect (AAE) were higher than for E- and N-residues. From residue interaction network analysis, we identified several local centrality measures and discussed their implications regarding the AAE. We also confirmed the selective cleavage of the backbone at D-proline bonds in proteins and further extend it to N-proline bonds. Finally, we note an enhancement of the AAE mechanism when the residue on the C-terminal side of D-, E- and N-residues is glycine. To the best of our knowledge, this is the first report of this phenomenon. Our study demonstrates the value of using statistical analyses of protein sequences and their predicted structures to better understand the fragmentation of the intact protein ions in the gas phase.

Locations and structures of influenza A virus packaging-associated signals and other functional elements via an in silico pipeline for predicting constrained features in RNA viruses.

Influenza A virus contains regions of its segmented genome associated with ability to package the segments into virions, but many such regions are poorly characterised. We provide detailed predictions of the key locations within these packaging-associated regions, and their structures, by applying a recently-improved pipeline for delineating constrained regions in RNA viruses and applying structural prediction algorithms. We find and characterise other known constrained regions within influenza A genomes, including the region associated with the PA-X frameshift, regions associated with alternative splicing, and constraint around the initiation motif for a truncated PB1 protein, PB1-N92, associated with avian viruses. We further predict the presence of constrained regions that have not previously been described. The extra characterisation our work provides allows investigation of these key regions for drug target potential, and points towards determinants of packaging compatibility between segments.

Computational analysis of affinity dynamics between the variants of SARS-CoV-2 spike protein (RBD) and human ACE-2 receptor

The novel coronavirus SARS-CoV-2 resulted in a significant worldwide health emergency known as the COVID-19 pandemic. This crisis has been marked by the widespread of various variants, with certain ones causing notable apprehension. In this study, we harnessed computational techniques to scrutinize these Variants of Concern (VOCs), including various Omicron subvariants. Our approach involved the use of protein structure prediction algorithms and molecular docking techniques, we have investigated the effects of mutations within the Receptor Binding Domain (RBD) of SARS-CoV-2 and how these mutations influence its interactions with the human angiotensin-converting enzyme 2 (hACE-2) receptor. Further we have predicted the structural alterations in the RBD of naturally occurring SARS-CoV-2 variants using the tr-Rosetta algorithm. Subsequent docking and binding analysis employing HADDOCK and PRODIGY illuminated crucial interactions occurring at the Receptor-Binding Motif (RBM). Our findings revealed a hierarchy of increased binding affinity between the human ACE2 receptor and the various RBDs, in the order of wild type (Wuhan-strain) < Beta < Alpha < Gamma < Omicron-B.1.1.529 < Delta < Omicron-BA.2.12.1 < Omicron-BA.5.2.1 < Omicron-BA.1.1. Notably, Omicron-BA.1.1 demonstrated the highest binding affinity of -17.4 kcal mol−1 to the hACE2 receptor when compared to all the mutant complexes. Additionally, our examination indicated that mutations occurring in active residues of the Receptor Binding Domain (RBD) consistently improved the binding affinity and intermolecular interactions in all mutant complexes. Analysis of the differences among variants has laid a foundation for the structure-based drug design targeting the RBD region of SARS-CoV-2.

Predicting structure performance of urban critical infrastructure: an augmented attention-based LSTM model

PurposeAccurate prediction of the structural condition of urban critical infrastructure is crucial for predictive maintenance. However, the existing prediction methods lack precision due to limitations in utilizing heterogeneous sensing data and domain knowledge as well as insufficient generalizability resulting from limited data samples. This paper integrates implicit and qualitative expert knowledge into quantifiable values in tunnel condition assessment and proposes a tunnel structure prediction algorithm that augments a state-of-the-art attention-based long short-term memory (LSTM) model with expert rating knowledge to achieve robust prediction results to reasonably allocate maintenance resources.Design/methodology/approachThrough formalizing domain experts' knowledge into quantitative tunnel condition index (TCI) with analytic hierarchy process (AHP), a fusion approach using sequence smoothing and sliding time window techniques is applied to the TCI and time-series sensing data. By incorporating both sensing data and expert ratings, an attention-based LSTM model is developed to improve prediction accuracy and reduce the uncertainty of structural influencing factors.FindingsThe empirical experiment in Dalian Road Tunnel in Shanghai, China showcases the effectiveness of the proposed method, which can comprehensively evaluate the tunnel structure condition and significantly improve prediction performance.Originality/valueThis study proposes a novel structure condition prediction algorithm that augments a state-of-the-art attention-based LSTM model with expert rating knowledge for robust prediction of structure condition of complex projects.

Tapcin, an In Vivo Active Dual Topoisomerase I/II Inhibitor Discovered by Synthetic Bioinformatic Natural Product (Syn-BNP)-Coupled Metagenomics.

DNA topoisomerases are attractive targets for anticancer agents. Dual topoisomerase I/II inhibitors are particularly appealing due to their reduced rates of resistance. A number of therapeutically relevant topoisomerase inhibitors are bacterial natural products. Mining the untapped chemical diversity encoded by soil microbiomes presents an opportunity to identify additional natural topoisomerase inhibitors. Here we couple metagenome mining, bioinformatic structure prediction algorithms, and chemical synthesis to produce the dual topoisomerase inhibitor tapcin. Tapcin is a mixed p-aminobenzoic acid (PABA)-thiazole with a rare tri-thiazole substructure and picomolar antiproliferative activity. Tapcin reduced colorectal adenocarcinoma HT-29 cell proliferation and tumor volume in mouse hollow fiber and xenograft models, respectively. In both studies it showed similar activity to the clinically used topoisomerase I inhibitor irinotecan. The study suggests that the interrogation of soil microbiomes using synthetic bioinformatic natural product methods has the potential to be a rewarding strategy for identifying potent, biomedically relevant, antiproliferative agents.

Tapcin, an In Vivo Active Dual Topoisomerase I/II Inhibitor Discovered by Synthetic Bioinformatic Natural Product (Syn‐BNP)‐Coupled Metagenomics

AbstractDNA topoisomerases are attractive targets for anticancer agents. Dual topoisomerase I/II inhibitors are particularly appealing due to their reduced rates of resistance. A number of therapeutically relevant topoisomerase inhibitors are bacterial natural products. Mining the untapped chemical diversity encoded by soil microbiomes presents an opportunity to identify additional natural topoisomerase inhibitors. Here we couple metagenome mining, bioinformatic structure prediction algorithms, and chemical synthesis to produce the dual topoisomerase inhibitor tapcin. Tapcin is a mixed p‐aminobenzoic acid (PABA)‐thiazole with a rare tri‐thiazole substructure and picomolar antiproliferative activity. Tapcin reduced colorectal adenocarcinoma HT‐29 cell proliferation and tumor volume in mouse hollow fiber and xenograft models, respectively. In both studies it showed similar activity to the clinically used topoisomerase I inhibitor irinotecan. The study suggests that the interrogation of soil microbiomes using synthetic bioinformatic natural product methods has the potential to be a rewarding strategy for identifying potent, biomedically relevant, antiproliferative agents.

SparseRNAfolD: optimized sparse RNA pseudoknot-free folding with dangle consideration

MotivationComputational RNA secondary structure prediction by free energy minimization is indispensable for analyzing structural RNAs and their interactions. These methods find the structure with the minimum free energy (MFE) among exponentially many possible structures and have a restrictive time and space complexity (O(n3)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$O(n^3)$$\\end{document} time and O(n2)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$O(n^2)$$\\end{document} space for pseudoknot-free structures) for longer RNA sequences. Furthermore, accurate free energy calculations, including dangle contributions can be difficult and costly to implement, particularly when optimizing for time and space requirements.ResultsHere we introduce a fast and efficient sparsified MFE pseudoknot-free structure prediction algorithm, SparseRNAFolD, that utilizes an accurate energy model that accounts for dangle contributions. While the sparsification technique was previously employed to improve the time and space complexity of a pseudoknot-free structure prediction method with a realistic energy model, SparseMFEFold, it was not extended to include dangle contributions due to the complexity of computation. This may come at the cost of prediction accuracy. In this work, we compare three different sparsified implementations for dangle contributions and provide pros and cons of each method. As well, we compare our algorithm to LinearFold, a linear time and space algorithm, where we find that in practice, SparseRNAFolD has lower memory consumption across all lengths of sequence and a faster time for lengths up to 1000 bases.ConclusionOur SparseRNAFolD algorithm is an MFE-based algorithm that guarantees optimality of result and employs the most general energy model, including dangle contributions. We provide a basis for applying dangles to sparsified recursion in a pseudoknot-free model that has the potential to be extended to pseudoknots.

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.

Limitations of Protein Structure Prediction Algorithms in Therapeutic Protein Development

The three-dimensional protein structure is pivotal in comprehending biological phenomena. It directly governs protein function and hence aids in drug discovery. The development of protein prediction algorithms, such as AlphaFold2, ESMFold, and trRosetta, has given much hope in expediting protein-based therapeutic discovery. Though no study has reported a conclusive application of these algorithms, the efforts continue with much optimism. We intended to test the application of these algorithms in rank-ordering therapeutic proteins for their instability during the pre-translational modification stages, as may be predicted according to the confidence of the structure predicted by these algorithms. The selected molecules were based on a harmonized category of licensed therapeutic proteins; out of the 204 licensed products, 188 that were not conjugated were chosen for analysis, resulting in a lack of correlation between the confidence scores and structural or protein properties. It is crucial to note here that the predictive accuracy of these algorithms is contingent upon the presence of the known structure of the protein in the accessible database. Consequently, our conclusion emphasizes that these algorithms primarily replicate information derived from existing structures. While our findings caution against relying on these algorithms for drug discovery purposes, we acknowledge the need for a nuanced interpretation. Considering their limitations and recognizing that their utility may be constrained to scenarios where known structures are available is important. Hence, caution is advised when applying these algorithms to characterize various attributes of therapeutic proteins without the support of adequate structural information. It is worth noting that the two main algorithms, AlfphaFold2 and ESMFold, also showed a 72% correlation in their scores, pointing to similar limitations. While much progress has been made in computational sciences, the Levinthal paradox remains unsolved.

Theoretical Investigation of Delafossite-Cu2ZnSnO4 as a Promising Photovoltaic Absorber.

In the quest for efficient and cost-effective photovoltaic absorber materials beyond silicon, considerable attention has been directed toward exploring alternatives. One such material, zincblende-derived Cu2ZnSnS4 (CZTS), has shown promise due to its ideal band gap size and high absorption coefficient. However, challenges such as structural defects and secondary phase formation have hindered its development. In this study, we examine the potential of another compound, Cu2ZnSnO4 (CZTO), with a similar composition to CZTS as a promising alternative. Employing ab initio density function theory (DFT) calculations in combination with an evolutionary structure prediction algorithm, we identify that the crystalline phase of delafossite structure is the most stable among the 900 (meta)stable CZTO. Its thermodynamic stability at room temperature is also confirmed by the molecular dynamics study. Excitingly, this new phase of CZTO displays a direct band gap where the dipole-allowed transition occurs, making it a strong candidate for efficient light absorptions. Furthermore, the estimation of spectroscopic limited maximum efficiency (SLME) directly demonstrates the high potential of delafossite-CZTO as a photovoltaic absorber. Our numerical results suggest that delafossite-CZTO holds promise for future photovoltaic applications.

Accelerating search for the polar phase stability of ferroelectric oxide by machine learning

Machine learning emerges to accelerate first-principles calculations in materials discovery and property prediction, but developing high-accuracy prediction models requires training on large datasets of relaxed structures, which can be computationally expensive. Besides, the crystal structure prediction algorithms show promising aspects but often deviate from the ground state, leading to poor predictions. Using a crystal graph convolutional neural network, we propose an alternative and efficient method of finding the ground state by utilizing the total energy of random unrelaxed structures. The prediction model is trained on a small dataset of unrelaxed energy obtained from first-principles calculations and effectively predicts the total energy of random structures with an accuracy almost equivalent to first-principles calculations. The resulting energy landscape is then reconstructed in terms of polarity. This leads to establishing an energy landscape of a perovskite oxide in a cost-effective way, enabling a polar phase stability map of strained BaTiO3, a representative ferroelectric oxide. The proposed methodology efficiently allows to obtain the ground state by eliminating the computationally intensive procedure of structure optimization, thus accelerating the ground state search to map out the phase stability of strained perovskite materials in a practical manner.

Selective Activation of a TRPC6 Ion Channel Over TRPC3 by Metalated Type-B Polycyclic Polyprenylated Acylphloroglucinols.

Selective modulation of TRPC6 ion channels is a promising therapeutic approach for neurodegenerative diseases and depression. A significant advancement showcases the selective activation of TRPC6 through metalated type-B PPAP, termed PPAP53. This success stems from PPAP53's 1,3-diketone motif facilitating metal coordination. PPAP53 is water-soluble and as potent as hyperforin, the gold standard in this field. In contrast to type-A, type-B PPAPs offer advantages such as gram-scale synthesis, easy derivatization, and long-term stability. Our investigations reveal PPAP53 selectively binding to the C-terminus of TRPC6. Although cryoelectron microscopy has resolved the majority of the TRPC6 structure, the binding site in the C-terminus remained unresolved. To address this issue, we employed state-of-the-art artificial-intelligence-based protein structure prediction algorithms to predict the missing region. Our computational results, validated against experimental data, indicate that PPAP53 binds to the 777LLKL780-region of the C-terminus, thus providing critical insights into the binding mechanism of PPAP53.

A machine-learning potential-based generative algorithm for on-lattice crystal structure prediction

A machine-learning potential-based generative algorithm for on-lattice crystal structure prediction

An Algorithm for the Development of a Recombinant Antiherpetic Subunit Vaccine Combining the Crystal Structure Analysis, AlphaFold2-Based Modeling, and Immunoinformatics

Using the envelope glycoprotein B (gB) crystal structure and digital prediction algorithm, the B- and T-cell antigenic determinants (epitopes) of human herpesvirus 1 (HHV-1), also known as herpes simplex virus 1 (HSV-1), were generated, and the method for their production in the form of recombinant proteins was proposed. First, the structure of the surface topological domain (ectodomain or spike) of gB with mapped epitopes was analyzed, and the most stable and immunogenic (due to their enrichment with B-and T-cell epitopes) subdomains were selected for the modeling of subunit vaccine prototypes using the AlphaFold2 (Google DeepMind, London, UK) artificial intelligence system. The proposed candidate vaccines included both small (about 100 amino acids) monomeric polypeptides, which were ideal for recombinant expression as fusion proteins, and a more complex polypeptide, which, due to its trimeric fold, looks like a miniature analog of the gB ectodomain. In this miniature analog, the ectodomain regions with the potential to interfere efficacious expression of soluble recombinant protein in Escherichia coli have been removed. The structural stability of the modeled proteins, confirmed by molecular dynamics simulation and host immune responses, predicted in silico, indicates the suitability of the two suggested polypeptides for generating subunit vaccines using recombinant DNA technology.

DNA Structure Design Is Improved Using an Artificially Expanded Alphabet of Base Pairs Including Loop and Mismatch Thermodynamic Parameters.

We show that in silico design of DNA secondary structures is improved by extending the base pairing alphabet beyond A-T and G-C to include the pair between 2-amino-8-(1'-β-d-2'-deoxyribofuranosyl)-imidazo-[1,2-a]-1,3,5-triazin-(8H)-4-one and 6-amino-3-(1'-β-d-2'-deoxyribofuranosyl)-5-nitro-(1H)-pyridin-2-one, abbreviated as P and Z. To obtain the thermodynamic parameters needed to include P-Z pairs in the designs, we performed 47 optical melting experiments and combined the results with previous work to fit free energy and enthalpy nearest neighbor folding parameters for P-Z pairs and G-Z wobble pairs. We find G-Z pairs have stability comparable to that of A-T pairs and should therefore be included as base pairs in structure prediction and design algorithms. Additionally, we extrapolated the set of loop, terminal mismatch, and dangling end parameters to include the P and Z nucleotides. These parameters were incorporated into the RNAstructure software package for secondary structure prediction and analysis. Using the RNAstructure Design program, we solved 99 of the 100 design problems posed by Eterna using the ACGT alphabet or supplementing it with P-Z pairs. Extending the alphabet reduced the propensity of sequences to fold into off-target structures, as evaluated by the normalized ensemble defect (NED). The NED values were improved relative to those from the Eterna example solutions in 91 of 99 cases in which Eterna-player solutions were provided. P-Z-containing designs had average NED values of 0.040, significantly below the 0.074 of standard-DNA-only designs, and inclusion of the P-Z pairs decreased the time needed to converge on a design. This work provides a sample pipeline for inclusion of any expanded alphabet nucleotides into prediction and design workflows.

LinearCoFold and LinearCoPartition: linear-time algorithms for secondary structure prediction of interacting RNA molecules.

Many RNAs function through RNA-RNA interactions. Fast and reliable RNA structure prediction with consideration of RNA-RNA interaction is useful, however, existing tools are either too simplistic or too slow. To address this issue, we present LinearCoFold, which approximates the complete minimum free energy structure of two strands in linear time, and LinearCoPartition, which approximates the cofolding partition function and base pairing probabilities in linear time. LinearCoFold and LinearCoPartition are orders of magnitude faster than RNAcofold. For example, on a sequence pair with combined length of 26,190nt, LinearCoFold is 86.8× faster than RNAcofold MFE mode, and LinearCoPartition is 642.3× faster than RNAcofold partition function mode. Surprisingly, LinearCoFold and LinearCoPartition's predictions have higher PPV and sensitivity of intermolecular base pairs. Furthermore, we apply LinearCoFold to predict the RNA-RNA interaction between SARS-CoV-2 genomic RNA (gRNA) and human U4 small nuclear RNA (snRNA), which has been experimentally studied, and observe that LinearCoFold's prediction correlates better with the wet lab results than RNAcofold's.

Hidden Glutathione Transferases in the Human Genome.

With the development of accurate protein structure prediction algorithms, artificial intelligence (AI) has emerged as a powerful tool in the field of structural biology. AI-based algorithms have been used to analyze large amounts of protein sequence data including the human proteome, complementing experimental structure data found in resources such as the Protein Data Bank. The EBI AlphaFold Protein Structure Database (for example) contains over 230 million structures. In this study, these data have been analyzed to find all human proteins containing (or predicted to contain) the cytosolic glutathione transferase (cGST) fold. A total of 39 proteins were found, including the alpha-, mu-, pi-, sigma-, zeta- and omega-class GSTs, intracellular chloride channels, metaxins, multisynthetase complex components, elongation factor 1 complex components and others. Three broad themes emerge: cGST domains as enzymes, as chloride ion channels and as protein-protein interaction mediators. As the majority of cGSTs are dimers, the AI-based structure prediction algorithm AlphaFold-multimer was used to predict structures of all pairwise combinations of these cGST domains. Potential homo- and heterodimers are described. Experimental biochemical and structure data is used to highlight the strengths and limitations of AI-predicted structures.

Advances in Peptide/Protein Structure Prediction Tools and their Relevance for Structural Biology in the Last Decade

Abstract: Peptides and proteins are involved in several biological processes at a molecular level. In this context, three-dimensional structure characterization and determination of peptides and proteins have helped researchers unravel the chemical and biological role of these macromolecules. Over 50 years, peptide and protein structures have been determined by experimental methods, including nuclear magnetic resonance (NMR), X-ray crystallography, and cryo-electron microscopy (cryo-EM). Therefore, an increasing number of atomic coordinates for peptides and proteins have been deposited in public databases, thus assisting the development of computational tools for predicting unknown 3D structures. In the last decade, a race for innovative methods has arisen in computational sciences, including more complex biological activity and structure prediction algorithms. As a result, peptide/protein theoretical models have achieved a new level of structure prediction accuracy compared with experimentally determined structures. Machine learning and deep learning approaches, for instance, incorporate fundamental aspects of peptide/protein geometry and include physical/biological knowledge about these macromolecules' experimental structures to build more precise computational models. Additionally, computational strategies have helped structural biology, including comparative, threading, and ab initio modeling and, more recently, prediction tools based on machine learning and deep learning. Bearing this in mind, here we provide a retrospective of protein and peptide structure prediction tools, highlighting their advances and obstacles and how they have assisted researchers in answering crucial biological questions.

Mono-valent salt corrections for RNA secondary structures in the ViennaRNA package

BackgroundRNA features a highly negatively charged phosphate backbone that attracts a cloud of counter-ions that reduce the electrostatic repulsion in a concentration dependent manner. Ion concentrations thus have a large influence on folding and stability of RNA structures. Despite their well-documented effects, salt effects are not handled consistently by currently available secondary structure prediction algorithms. Combining Debye-Hückel potentials for line charges and Manning’s counter-ion condensation theory, Einert et al. (Biophys J 100: 2745-2753, 2011) modeled the energetic contributions of monovalent cations on loops and helices.ResultsThe model of Einert et al. is adapted to match the structure of the dynamic programming recursion of RNA secondary structure prediction algorithms. An empirical term describing the salt dependence of the duplex initiation energy is added to improve co-folding predictions for two or more RNA strands. The slightly modified model is implemented in the ViennaRNA package in such way that only the energy parameters but not the algorithmic structure is affected. A comparison with data from the literature show that predicted free energies and melting temperatures are in reasonable agreement with experiments.ConclusionThe new feature in the ViennaRNA package makes it possible to study effects of salt concentrations on RNA folding in a systematic manner. Strictly speaking, the model pertains only to mono-valent cations, and thus covers the most important parameter, i.e., the NaCl concentration. It remains a question for future research to what extent unspecific effects of bi- and tri-valent cations can be approximated in a similar manner.AvailabilityCorrections for the concentration of monovalent cations are available in the ViennaRNA package starting from version 2.6.0.

Deep Learning-Based Prediction of Contact Maps and Crystal Structures of Inorganic Materials.

Crystal structure prediction is one of the major unsolved problems in materials science. Traditionally, this problem is formulated as a global optimization problem for which global search algorithms are combined with first-principles free energy calculations to predict the ground-state crystal structure of a given material composition. These ab initio algorithms are currently too slow for predicting complex material structures. Inspired by the AlphaFold algorithm for protein structure prediction, herein, we propose AlphaCrystal, a crystal structure prediction algorithm that combines a deep residual neural network model for predicting the atomic contact map of a target material followed by three-dimensional (3D) structure reconstruction using genetic algorithms. Extensive experiments on 20 benchmark structures showed that our AlphaCrystal algorithm can predict structures close to the ground truth structures, which can significantly speed up the crystal structure prediction and handle relatively large systems.