Zika Virus Infection Induces Interleukin-1ß-Mediated Inflammatory Responses by Macrophages in the Brain of an Adult Mouse Model.

During the 2015-2016 Zika virus (ZIKV) epidemic, ZIKV-associated neurological diseases were reported in adults, including microcephaly, Guillain-Barre syndrome, myelitis, meningoencephalitis, and fatal encephalitis. However, the mechanisms underlying the neuropathogenesis of ZIKV infection are not yet fully understood. In this study, we used an adult ZIKV infection mouse model (Ifnar1-/-) to investigate the mechanisms underlying neuroinflammation and neuropathogenesis. ZIKV infection induced the expression of proinflammatory cytokines, including interleukin-1ß (IL-1ß), IL-6, gamma interferon, and tumor necrosis factor alpha, in the brains of Ifnar1-/- mice. RNA-seq analysis of the infected mouse brain also revealed that genes involved in innate immune responses and cytokine-mediated signaling pathways were significantly upregulated at 6 days postinfection. Furthermore, ZIKV infection induced macrophage infiltration and activation and augmented IL-1ß expression, whereas microgliosis was not observed in the brain. Using human monocyte THP-1 cells, we confirmed that ZIKV infection promotes inflammatory cell death and increases IL-1ß secretion. In addition, expression of the complement component C3, which is associated with neurodegenerative diseases and known to be upregulated by proinflammatory cytokines, was induced by ZIKV infection through the IL-1ß-mediated pathway. An increase in C5a produced by complement activation in the brains of ZIKV-infected mice was also verified. Taken together, our results suggest that ZIKV infection in the brain of this animal model augments IL-1ß expression in infiltrating macrophages and elicits IL-1ß-mediated inflammation, which can lead to the destructive consequences of neuroinflammation. IMPORTANCE Zika virus (ZIKV) associated neurological impairments are an important global health problem. Our results suggest that ZIKV infection in the mouse brain can induce IL-1ß-mediated inflammation and complement activation, thereby contributing to the development of neurological disorders. Thus, our findings reveal a mechanism by which ZIKV induces neuroinflammation in the mouse brain. Although we used adult type I interferon receptor IFNAR knockout (Ifnar1-/-) mice owing to the limited mouse models of ZIKV pathogenesis, our conclusions contributed to the understanding ZIKV-associated neurological diseases to develop treatment strategies for patients with ZIKV infection based on these findings.

Discovery of thiophen-2-ylmethylene bis-dimedone derivatives as novel WRN inhibitors for treating cancers with microsatellite instability.

Microsatellite instability (MSI) is a hypermutable condition caused by DNA mismatch repair system defects, contributing to the development of various cancer types. Recent research has identified Werner syndrome ATP-dependent helicase (WRN) as a promising synthetic lethal target for MSI cancers. Herein, we report the first discovery of thiophen-2-ylmethylene bis-dimedone derivatives as novel WRN inhibitors for MSI cancer therapy. Initial computational analysis and biological evaluation identified a new scaffold for a WRN inhibitor. Subsequent SAR study led to the discovery of a highly potent WRN inhibitor. Furthermore, we demonstrated that the optimal compound induced DNA damage and apoptotic cell death in MSI cancer cells by inhibiting WRN. This study provides a new pharmacophore for WRN inhibitors, emphasizing their therapeutic potential for MSI cancers.

A-Prot: protein structure modeling using MSA transformer.

BACKGROUND: The accuracy of protein 3D structure prediction has been dramatically improved with the help of advances in deep learning. In the recent CASP14, Deepmind demonstrated that their new version of AlphaFold (AF) produces highly accurate 3D models almost close to experimental structures. The success of AF shows that the multiple sequence alignment of a sequence contains rich evolutionary information, leading to accurate 3D models. Despite the success of AF, only the prediction code is open, and training a similar model requires a vast amount of computational resources. Thus, developing a lighter prediction model is still necessary. RESULTS: In this study, we propose a new protein 3D structure modeling method, A-Prot, using MSA Transformer, one of the state-of-the-art protein language models. An MSA feature tensor and row attention maps are extracted and converted into 2D residue-residue distance and dihedral angle predictions for a given MSA. We demonstrated that A-Prot predicts long-range contacts better than the existing methods. Additionally, we modeled the 3D structures of the free modeling and hard template-based modeling targets of CASP14. The assessment shows that the A-Prot models are more accurate than most top server groups of CASP14. CONCLUSION: These results imply that A-Prot accurately captures the evolutionary and structural information of proteins with relatively low computational cost. Thus, A-Prot can provide a clue for the development of other protein property prediction methods.

AK-Score: Accurate Protein-Ligand Binding Affinity Prediction Using an Ensemble of 3D-Convolutional Neural Networks.

Accurate prediction of the binding affinity of a protein-ligand complex is essential for efficient and successful rational drug design. Therefore, many binding affinity prediction methods have been developed. In recent years, since deep learning technology has become powerful, it is also implemented to predict affinity. In this work, a new neural network model that predicts the binding affinity of a protein-ligand complex structure is developed. Our model predicts the binding affinity of a complex using the ensemble of multiple independently trained networks that consist of multiple channels of 3-D convolutional neural network layers. Our model was trained using the 3772 protein-ligand complexes from the refined set of the PDBbind-2016 database and tested using the core set of 285 complexes. The benchmark results show that the Pearson correlation coefficient between the predicted binding affinities by our model and the experimental data is 0.827, which is higher than the state-of-the-art binding affinity prediction scoring functions. Additionally, our method ranks the relative binding affinities of possible multiple binders of a protein quite accurately, comparable to the other scoring functions. Last, we measured which structural information is critical for predicting binding affinity and found that the complementarity between the protein and ligand is most important.

Comparative transcriptome analysis identified candidate genes involved in mycelium browning in Lentinula edodes.

BACKGROUND: Lentinula edodes is one of the most popular edible mushroom species in the world and contains useful medicinal components, such as lentinan. The light-induced formation of brown film on the vegetative mycelial tissues of L. edodes is an important process for ensuring the quantity and quality of this edible mushroom. To understand the molecular mechanisms underlying this critical developmental process in L. edodes, we characterized the morphological phenotypic changes in a strain, Chamaram, associated with abnormal brown film formation and compared its genome-wide transcriptional features. RESULTS: In the present study, we performed genome-wide transcriptome analyses of different vegetative mycelium growth phenotypes, namely, early white, normal brown, and defective dark yellow partial brown films phenotypes which were exposed to different light conditions. The analysis revealed the identification of clusters of genes specific to the light-induced brown film phenotypes. These genes were significantly associated with light sensing via photoreceptors such as FMN- and FAD-bindings, signal transduction by kinases and GPCRs, melanogenesis via activation of tyrosinases, and cell wall degradation by glucanases, chitinases, and laccases, which suggests these processes are involved in the formation of mycelial browning in L. edodes. Interestingly, hydrophobin genes such as SC1 and SC3 exhibited divergent expression levels in the normal and abnormal brown mycelial films, indicating the ability of these genes to act in fruiting body initiation and formation of dikaryotic mycelia. Furthermore, we identified the up-regulation of glycoside hydrolase domain-containing genes in the normal brown film but not in the abnormal film phenotype, suggesting that cell wall degradation in the normal brown film phenotype is crucial in the developmental processes related to the initiation and formation of fruiting bodies. CONCLUSIONS: This study systematically analysed the expression patterns of light-induced browning-related genes in L. edodes. Our findings provide information for further investigations of browning formation mechanisms in L. edodes and a foundation for future L. edodes breeding.

Whole genome sequencing of an ethnic Pathan (Pakhtun) from the north-west of Pakistan.

BACKGROUND: Pakistan covers a key geographic area in human history, being both part of the Indus River region that acted as one of the cradles of civilization and as a link between Western Eurasia and Eastern Asia. This region is inhabited by a number of distinct ethnic groups, the largest being the Punjabi, Pathan (Pakhtuns), Sindhi, and Baloch. RESULTS: We analyzed the first ethnic male Pathan genome by sequencing it to 29.7-fold coverage using the Illumina HiSeq2000 platform. A total of 3.8 million single nucleotide variations (SNVs) and 0.5 million small indels were identified by comparing with the human reference genome. Among the SNVs, 129,441 were novel, and 10,315 nonsynonymous SNVs were found in 5,344 genes. SNVs were annotated for health consequences and high risk diseases, as well as possible influences on drug efficacy. We confirmed that the Pathan genome presented here is representative of this ethnic group by comparing it to a panel of Central Asians from the HGDP-CEPH panels typed for ~650 k SNPs. The mtDNA (H2) and Y haplogroup (L1) of this individual were also typical of his geographic region of origin. Finally, we reconstruct the demographic history by PSMC, which highlights a recent increase in effective population size compatible with admixture between European and Asian lineages expected in this geographic region. CONCLUSIONS: We present a whole-genome sequence and analyses of an ethnic Pathan from the north-west province of Pakistan. It is a useful resource to understand genetic variation and human migration across the whole Asian continent.

Blind prediction of interfacial water positions in CAPRI.

We report the first assessment of blind predictions of water positions at protein-protein interfaces, performed as part of the critical assessment of predicted interactions (CAPRI) community-wide experiment. Groups submitting docking predictions for the complex of the DNase domain of colicin E2 and Im2 immunity protein (CAPRI Target 47), were invited to predict the positions of interfacial water molecules using the method of their choice. The predictions-20 groups submitted a total of 195 models-were assessed by measuring the recall fraction of water-mediated protein contacts. Of the 176 high- or medium-quality docking models-a very good docking performance per se-only 44% had a recall fraction above 0.3, and a mere 6% above 0.5. The actual water positions were in general predicted to an accuracy level no better than 1.5 Å, and even in good models about half of the contacts represented false positives. This notwithstanding, three hotspot interface water positions were quite well predicted, and so was one of the water positions that is believed to stabilize the loop that confers specificity in these complexes. Overall the best interface water predictions was achieved by groups that also produced high-quality docking models, indicating that accurate modelling of the protein portion is a determinant factor. The use of established molecular mechanics force fields, coupled to sampling and optimization procedures also seemed to confer an advantage. Insights gained from this analysis should help improve the prediction of protein-water interactions and their role in stabilizing protein complexes.

Genome-wide analysis of DNA methylation patterns in horse.

BACKGROUND: DNA methylation is an epigenetic regulatory mechanism that plays an essential role in mediating biological processes and determining phenotypic plasticity in organisms. Although the horse reference genome and whole transcriptome data are publically available the global DNA methylation data are yet to be known. RESULTS: We report the first genome-wide DNA methylation characteristics data from skeletal muscle, heart, lung, and cerebrum tissues of thoroughbred (TH) and Jeju (JH) horses, an indigenous Korea breed, respectively by methyl-DNA immunoprecipitation sequencing. The analysis of the DNA methylation patterns indicated that the average methylation density was the lowest in the promoter region, while the density in the coding DNA sequence region was the highest. Among repeat elements, a relatively high density of methylation was observed in long interspersed nuclear elements compared to short interspersed nuclear elements or long terminal repeat elements. We also successfully identified differential methylated regions through a comparative analysis of corresponding tissues from TH and JH, indicating that the gene body regions showed a high methylation density. CONCLUSIONS: We provide report the first DNA methylation landscape and differentially methylated genomic regions (DMRs) of thoroughbred and Jeju horses, providing comprehensive DMRs maps of the DNA methylome. These data are invaluable resource to better understanding of epigenetics in the horse providing information for the further biological function analyses.

GalaxyGemini: a web server for protein homo-oligomer structure prediction based on similarity.

SUMMARY: A large number of proteins function as homo-oligomers; therefore, predicting homo-oligomeric structure of proteins is of primary importance for understanding protein function at the molecular level. Here, we introduce a web server for prediction of protein homo-oligomer structure. The server takes a protein monomer structure as input and predicts its homo-oligomer structure from oligomer templates selected based on sequence and tertiary/quaternary structure similarity. Using protein model structures as input, the server shows clear improvement over the best methods of CASP9 in predicting oligomeric structures from amino acid sequences. AVAILABILITY: http://galaxy.seoklab.org/gemini. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

GalaxyWEB server for protein structure prediction and refinement.

Three-dimensional protein structures provide invaluable information for understanding and regulating biological functions of proteins. The GalaxyWEB server predicts protein structure from sequence by template-based modeling and refines loop or terminus regions by ab initio modeling. This web server is based on the method tested in CASP9 (9th Critical Assessment of techniques for protein Structure Prediction) as 'Seok-server', which was assessed to be among top performing template-based modeling servers. The method generates reliable core structures from multiple templates and re-builds unreliable loops or termini by using an optimization-based refinement method. In addition to structure prediction, a user can also submit a refinement only job by providing a starting model structure and locations of loops or termini to refine. The web server can be freely accessed at http://galaxy.seoklab.org/.

The FALC-Loop web server for protein loop modeling.

The FALC-Loop web server provides an online interface for protein loop modeling by employing an ab initio loop modeling method called FALC (fragment assembly and analytical loop closure). The server may be used to construct loop regions in homology modeling, to refine unreliable loop regions in experimental structures or to model segments of designed sequences. The FALC method is computationally less expensive than typical ab initio methods because the conformational search space is effectively reduced by the use of fragments derived from a structure database. The analytical loop closure algorithm allows efficient search for loop conformations that fit into the protein framework starting from the fragment-assembled structures. The FALC method shows prediction accuracy comparable to other state-of-the-art loop modeling methods. Top-ranked model structures can be visualized on the web server, and an ensemble of loop structures can be downloaded for further analysis. The web server can be freely accessed at http://falc-loop.seoklab.org/.

Intercellular cross-talk through lineage-specific gap junction of cancer-associated fibroblasts related to stromal fibrosis and prognosis.

Stromal fibrosis in cancer is usually associated with poor prognosis and chemotherapy resistance. It is thought to be caused by fibroblasts; however, the exact mechanism is not yet well understood. The study aimed to identify lineage-specific cancer-associated fibroblast (CAF) subgroup and their associations with extracellular matrix remodeling and clinical significances in various tumor types using single-cell and bulk RNA sequencing data. Through unsupervised clustering, six subclusters of CAFs were identified, including a cluster with exclusively high gap junction protein beta-2 (GJB2) expression. This cluster was named GJB2-positive CAF. It was found to be a unique subgroup of terminally differentiated CAFs associated with collagen gene expression and extracellular matrix remodeling. GJB2-positive CAFs showed higher communication frequency with vascular endothelial cells and cancer cells than GJB2-negative CAFs. Moreover, GJB2 was poorly expressed in normal tissues, indicating that its expression is dependent on interaction with other cells, including vascular endothelial cells and cancer cells. Finally, the study investigated the clinical significance of GJB2 signature score for GJB2-positive CAFs in cancer and found a correlation with poor prognosis. These results suggest that GJB2-positive CAF is a unique fibroblast subtype involved in extracellular matrix remodeling, with significant clinical implications in cancer.

GalaxyTBM: template-based modeling by building a reliable core and refining unreliable local regions.

BACKGROUND: Protein structures can be reliably predicted by template-based modeling (TBM) when experimental structures of homologous proteins are available. However, it is challenging to obtain structures more accurate than the single best templates by either combining information from multiple templates or by modeling regions that vary among templates or are not covered by any templates. RESULTS: We introduce GalaxyTBM, a new TBM method in which the more reliable core region is modeled first from multiple templates and less reliable, variable local regions, such as loops or termini, are then detected and re-modeled by an ab initio method. This TBM method is based on "Seok-server," which was tested in CASP9 and assessed to be amongst the top TBM servers. The accuracy of the initial core modeling is enhanced by focusing on more conserved regions in the multiple-template selection and multiple sequence alignment stages. Additional improvement is achieved by ab initio modeling of up to 3 unreliable local regions in the fixed framework of the core structure. Overall, GalaxyTBM reproduced the performance of Seok-server, with GalaxyTBM and Seok-server resulting in average GDT-TS of 68.1 and 68.4, respectively, when tested on 68 single-domain CASP9 TBM targets. For application to multi-domain proteins, GalaxyTBM must be combined with domain-splitting methods. CONCLUSION: Application of GalaxyTBM to CASP9 targets demonstrates that accurate protein structure prediction is possible by use of a multiple-template-based approach, and ab initio modeling of variable regions can further enhance the model quality.

Whole transcriptome analyses of six thoroughbred horses before and after exercise using RNA-Seq.

BACKGROUND: Thoroughbred horses are the most expensive domestic animals, and their running ability and knowledge about their muscle-related diseases are important in animal genetics. While the horse reference genome is available, there has been no large-scale functional annotation of the genome using expressed genes derived from transcriptomes. RESULTS: We present a large-scale analysis of whole transcriptome data. We sequenced the whole mRNA from the blood and muscle tissues of six thoroughbred horses before and after exercise. By comparing current genome annotations, we identified 32,361 unigene clusters spanning 51.83 Mb that contained 11,933 (36.87%) annotated genes. More than 60% (20,428) of the unigene clusters did not match any current equine gene model. We also identified 189,973 single nucleotide variations (SNVs) from the sequences aligned against the horse reference genome. Most SNVs (171,558 SNVs; 90.31%) were novel when compared with over 1.1 million equine SNPs from two SNP databases. Using differential expression analysis, we further identified a number of exercise-regulated genes: 62 up-regulated and 80 down-regulated genes in the blood, and 878 up-regulated and 285 down-regulated genes in the muscle. Six of 28 previously-known exercise-related genes were over-expressed in the muscle after exercise. Among the differentially expressed genes, there were 91 transcription factor-encoding genes, which included 56 functionally unknown transcription factor candidates that are probably associated with an early regulatory exercise mechanism. In addition, we found interesting RNA expression patterns where different alternative splicing forms of the same gene showed reversed expressions before and after exercising. CONCLUSION: The first sequencing-based horse transcriptome data, extensive analyses results, deferentially expressed genes before and after exercise, and candidate genes that are related to the exercise are provided in this study.

Retrosynthetic reaction pathway prediction through neural machine translation of atomic environments.

Designing efficient synthetic routes for a target molecule remains a major challenge in organic synthesis. Atom environments are ideal, stand-alone, chemically meaningful building blocks providing a high-resolution molecular representation. Our approach mimics chemical reasoning, and predicts reactant candidates by learning the changes of atom environments associated with the chemical reaction. Through careful inspection of reactant candidates, we demonstrate atom environments as promising descriptors for studying reaction route prediction and discovery. Here, we present a new single-step retrosynthesis prediction method, viz. RetroTRAE, being free from all SMILES-based translation issues, yields a top-1 accuracy of 58.3% on the USPTO test dataset, and top-1 accuracy reaches to 61.6% with the inclusion of highly similar analogs, outperforming other state-of-the-art neural machine translation-based methods. Our methodology introduces a novel scheme for fragmental and topological descriptors to be used as natural inputs for retrosynthetic prediction tasks.

S-Pred: protein structural property prediction using MSA transformer.

Predicting the local structural features of a protein from its amino acid sequence helps its function prediction to be revealed and assists in three-dimensional structural modeling. As the sequence-structure gap increases, prediction methods have been developed to bridge this gap. Additionally, as the size of the structural database and computing power increase, the performance of these methods have also significantly improved. Herein, we present a powerful new tool called S-Pred, which can predict eight-state secondary structures (SS8), accessible surface areas (ASAs), and intrinsically disordered regions (IDRs) from a given sequence. For feature prediction, S-Pred uses multiple sequence alignment (MSA) of a query sequence as an input. The MSA input is converted to features by the MSA Transformer, which is a protein language model that uses an attention mechanism. A long short-term memory (LSTM) was employed to produce the final prediction. The performance of S-Pred was evaluated on several test sets, and the program consistently provided accurate predictions. The accuracy of the SS8 prediction was approximately 76%, and the Pearson's correlation between the experimental and predicted ASAs was 0.84. Additionally, an IDR could be accurately predicted with an F1-score of 0.514. The program is freely available at https://github.com/arontier/S_Pred_Paper and https://ad3.io as a code and a web server.

SARS-CoV-2 Infection of Microglia Elicits Proinflammatory Activation and Apoptotic Cell Death.

Accumulating evidence suggests that severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection causes various neurological symptoms in patients with coronavirus disease 2019 (COVID-19). The most dominant immune cells in the brain are microglia. Yet, the relationship between neurological manifestations, neuroinflammation, and host immune response of microglia to SARS-CoV-2 has not been well characterized. Here, we reported that SARS-CoV-2 can directly infect human microglia, eliciting M1-like proinflammatory responses, followed by cytopathic effects. Specifically, SARS-CoV-2 infected human microglial clone 3 (HMC3), leading to inflammatory activation and cell death. RNA sequencing (RNA-seq) analysis also revealed that endoplasmic reticulum (ER) stress and immune responses were induced in the early, and apoptotic processes in the late phases of viral infection. SARS-CoV-2-infected HMC3 showed the M1 phenotype and produced proinflammatory cytokines, such as interleukin (IL)-1ß, IL-6, and tumor necrosis factor α (TNF-α), but not the anti-inflammatory cytokine IL-10. After this proinflammatory activation, SARS-CoV-2 infection promoted both intrinsic and extrinsic death receptor-mediated apoptosis in HMC3. Using K18-hACE2 transgenic mice, murine microglia were also infected by intranasal inoculation of SARS-CoV-2. This infection induced the acute production of proinflammatory microglial IL-6 and TNF-α and provoked a chronic loss of microglia. Our findings suggest that microglia are potential mediators of SARS-CoV-2-induced neurological problems and, consequently, can be targets of therapeutic strategies against neurological diseases in patients with COVID-19. IMPORTANCE Recent studies reported neurological and cognitive sequelae in patients with COVID-19 months after the viral infection with several symptoms, including ageusia, anosmia, asthenia, headache, and brain fog. Our conclusions raise awareness of COVID-19-related microglia-mediated neurological disorders to develop treatment strategies for the affected patients. We also indicated that HMC3 was a novel human cell line susceptible to SARS-CoV-2 infection that exhibited cytopathic effects, which could be further used to investigate cellular and molecular mechanisms of neurological manifestations of patients with COVID-19.

Refinement of protein termini in template-based modeling using conformational space annealing.

The rapid increase in the number of experimentally determined protein structures in recent years enables us to obtain more reliable protein tertiary structure models than ever by template-based modeling. However, refinement of template-based models beyond the limit available from the best templates is still needed for understanding protein function in atomic detail. In this work, we develop a new method for protein terminus modeling that can be applied to refinement of models with unreliable terminus structures. The energy function for terminus modeling consists of both physics-based and knowledge-based potential terms with carefully optimized relative weights. Effective sampling of both the framework and terminus is performed using the conformational space annealing technique. This method has been tested on a set of termini derived from a nonredundant structure database and two sets of termini from the CASP8 targets. The performance of the terminus modeling method is significantly improved over our previous method that does not employ terminus refinement. It is also comparable or superior to the best server methods tested in CASP8. The success of the current approach suggests that similar strategy may be applied to other types of refinement problems such as loop modeling or secondary structure rearrangement.

LigDockCSA: protein-ligand docking using conformational space annealing.

Protein-ligand docking techniques are one of the essential tools for structure-based drug design. Two major components of a successful docking program are an efficient search method and an accurate scoring function. In this work, a new docking method called LigDockCSA is developed by using a powerful global optimization technique, conformational space annealing (CSA), and a scoring function that combines the AutoDock energy and the piecewise linear potential (PLP) torsion energy. It is shown that the CSA search method can find lower energy binding poses than the Lamarckian genetic algorithm of AutoDock. However, lower-energy solutions CSA produced with the AutoDock energy were often less native-like. The loophole in the AutoDock energy was fixed by adding a torsional energy term, and the CSA search on the refined energy function is shown to improve the docking performance. The performance of LigDockCSA was tested on the Astex diverse set which consists of 85 protein-ligand complexes. LigDockCSA finds the best scoring poses within 2 Å root-mean-square deviation (RMSD) from the native structures for 84.7% of the test cases, compared to 81.7% for AutoDock and 80.5% for GOLD. The results improve further to 89.4% by incorporating the conformational entropy.

Substructure-based neural machine translation for retrosynthetic prediction.

With the rapid improvement of machine translation approaches, neural machine translation has started to play an important role in retrosynthesis planning, which finds reasonable synthetic pathways for a target molecule. Previous studies showed that utilizing the sequence-to-sequence frameworks of neural machine translation is a promising approach to tackle the retrosynthetic planning problem. In this work, we recast the retrosynthetic planning problem as a language translation problem using a template-free sequence-to-sequence model. The model is trained in an end-to-end and a fully data-driven fashion. Unlike previous models translating the SMILES strings of reactants and products, we introduced a new way of representing a chemical reaction based on molecular fragments. It is demonstrated that the new approach yields better prediction results than current state-of-the-art computational methods. The new approach resolves the major drawbacks of existing retrosynthetic methods such as generating invalid SMILES strings. Specifically, our approach predicts highly similar reactant molecules with an accuracy of 57.7%. In addition, our method yields more robust predictions than existing methods.