The sarbecovirus subgenus of betacoronaviruses is widely distributed throughout bats and other mammals globally and includes human pathogens, SARS-CoV and SARS-CoV-2. The most studied sarbecoviruses use the host protein, ACE2, to infect cells. Curiously, the majority of sarbecoviruses identified to date do not use ACE2 and cannot readily acquire ACE2 binding through point mutations. We previously screened a broad panel of sarbecovirus spikes for cell entry and observed bat-derived viruses that could infect human cells, independent of ACE2. Here we further investigate the sequence determinants of cell entry for ACE2-independent bat sarbecoviruses. We employed a network science-based approach to visualize sequence and entry phenotype similarities across the diversity of sarbecovirus spike protein sequences. We then verified these computational results and mapped determinants of viral entry into human cells using recombinant chimeric spike proteins within an established viral pseudotype assay. We show ACE2-independent viruses that can infect human and bat cells in culture have a similar putative receptor binding motif, which can impart human cell entry into other bat sarbecovirus spikes that cannot otherwise infect human cells. These sequence determinants of human cell entry map to a surface-exposed protrusion from the predicted bat sarbecovirus spike receptor binding domain structure. Our findings provide further evidence of a group of bat-derived sarbecoviruses with zoonotic potential and demonstrate the utility in applying network science to phenotypic mapping and prediction. This work was supported by Washington State University and the Paul G. Allen School for Global Health.

Reconstructing and visualizing phylogenetic relationships among living organisms is a fundamental challenge because not all organisms share the same genes. As a result, the first phylogenetic visualizations employed a single gene, e.g., rRNA genes, sufficiently conserved to be present in all organisms but divergent enough to provide discrimination between groups. As more genome data became available, researchers began concatenating different combinations of genes or proteins to construct phylogenetic trees believed to be more robust because they incorporated more information. However, the genes or proteins chosen were based on ad hoc approaches. The large number of complete genome sequences available today allows the use of whole genomes to analyze relationships among organisms rather than using an ad hoc set of genes. We present a systematic approach for constructing a phylogenetic tree based on simultaneously clustering the complete proteomes of 360 bacterial species. From the homologous clusters, we identify 49 protein sequences shared by 99% of the organisms to build a tree. Of the 49 sequences, 47 have homologous sequences in both archaea and eukarya. The clusters are also used to create a network from which bacterial species with horizontally-transferred genes from other phyla are identified.

We clustered 8.76 M protein sequences deduced from 2,307 completely sequenced Proteobacterial genomes resulting in 707,311 clusters of one or more sequences of which 224,442 ranged in size from 2 to 2,894 sequences. To our knowledge this is the first study of this scale. We were surprised to find that no single cluster contained a representative sequence from all the organisms in the study. Given the minimal genome concept, we expected to find a shared set of proteins. To determine why the clusters did not have universal representation we chose four essential proteins, the chaperonin GroEL, DNA dependent RNA polymerase subunits beta and beta′ (RpoB/RpoB′), and DNA polymerase I (PolA), representing fundamental cellular functions, and examined their cluster distribution. We found these proteins to be remarkably conserved with certain caveats. Although the groEL gene was universally conserved in all the organisms in the study, the protein was not represented in all the deduced proteomes. The genes for RpoB and RpoB′ were missing from two genomes and merged in 88, and the sequences were sufficiently divergent that they formed separate clusters for 18 RpoB proteins (seven clusters) and 14 RpoB′ proteins (three clusters). For PolA, 52 organisms lacked an identifiable sequence, and seven sequences were sufficiently divergent that they formed five separate clusters. Interestingly, organisms lacking an identifiable PolA and those with divergent RpoB/RpoB′ were predominantly endosymbionts. Furthermore, we present a range of examples of annotation issues that caused the deduced proteins to be incorrectly represented in the proteome. These annotation issues made our task of determining protein conservation more difficult than expected and also represent a significant obstacle for high-throughput analyses.

Advances in genome sequencing technology and computing power have brought about the explosive growth of sequenced genomes in public repositories with a concomitant increase in annotation errors. Many protein sequences are annotated using computational analysis rather than experimental verification, leading to inaccuracies in annotation. Confirmation of existing protein annotations is urgently needed before misannotation becomes even more prevalent due to error propagation. In this work we present a novel approach for automatically confirming the existence of manually curated information with experimental evidence of protein annotation. Our ensemble learning method uses a combination of recurrent convolutional neural network, logistic regression, and support vector machine models. Natural language processing in the form of word embeddings is used with journal publication titles retrieved from the UniProtKB database. Importantly, we use recall as our most significant metric to ensure the maximum number of verifications possible; results are reported to a human curator for confirmation. Our ensemble model achieves 91.25% recall, 71.26% accuracy, 65.19% precision, and an F1 score of 76.05% and outperforms the Bidirectional Encoder Representations from Transformers for Biomedical Text Mining (BioBERT) model with fine-tuning using the same data.