Integrating gene annotation with orthology inference at scale

In the paper

“Distinguishing orthologs and paralogs is a fundamental problem in evolutionary and molecular biology (1) and a prerequisite for many genomic analyses, including reconstructing phylogenetic trees, predicting gene function, investigating molecular and genome evolution, and discovering differences in genes that underlie phenotypes of the sequenced species (2–6).”

What the sources say

• Supported

  Citation (1) supports the first clause, not the later list of genomic analyses cited to (2–6). The Conclusions call orthology and paralogy central to comparative and functional genomics, and the Gene homology relationships section explains why distinguishing them can be complex. (LLM, conf=0.90)

  cites [1] Gabaldon T, Koonin EV, Functional and evolutionary implications of gene orthology. Nature reviews. Genetics 14, 360– 366 (2013). PMC5877793 23552219 source
• ?Not checked — no full text of the sourcecites [2] Kapli P, Yang Z, Telford MJ, Phylogenetic tree building in the genomic age. Nature reviews. Genetics 21, 428– 444 (2020). 32424311 source
• Supported

  The citation is attached to the claim that distinguishing orthologs from paralogs is fundamental, not to the later list of genomic analyses cited to other sources. The Discussion calls the distinction a central concept of phylogenomics, and the Introduction describes it as a canonical refinement used in relating gene evolution to function. (LLM, conf=0.88)

  cites [3] Altenhoff AM, Studer RA, Robinson-Rechavi M, Dessimoz C, Resolving the ortholog conjecture: orthologs tend to be weakly, but significantly, more similar in function than paralogs. PLoS computational biology 8, e1002514 (2012). 22615551 PMC3355068 source
• Partly supported

  The Introduction (page unavailable) calls identifying orthologs a long-standing evolutionary problem and explains its importance for predicting gene function. The Conclusions (page unavailable) also describe orthology-based functional annotation of genomes. The paper does not establish the broader claim that distinguishing orthologs and paralogs is a prerequisite for the other listed analyses, such as reconstructing phylogenetic trees or identifying phenotype-related gene differences. (LLM, conf=0.83)

  cites [4] Huerta-Cepas J, Fast Genome-Wide Functional Annotation through Orthology Assignment by eggNOG-Mapper. Molecular biology and evolution 34, 2115– 2122 (2017). 28460117 PMC5850834 sourceExperimental — accuracy not yet measured
• Supported

  Citation (2–6) supports the sentence’s point about genomic analyses, rather than the separately cited claim that distinguishing orthologs and paralogs is a fundamental problem. The Methods section ('Investigating the role of gene loss for mammalian adaptations'; page unavailable) says the study used 1:1 orthologous genes to connect gene losses with phenotypic adaptations, making this paper an example of the relevant analysis. (LLM, conf=0.85)

  cites [5] Sharma V, A genomics approach reveals insights into the importance of gene losses for mammalian adaptations. Nature communications 9, 1215 (2018). PMC5865188 29572503 source
• Partly supported

  The citation is attached to the first clause, which calls distinguishing orthologs from paralogs a fundamental problem. In Methods > Phylogenomics (page unavailable), the paper uses genes classified as 1:1 orthologues to infer a species tree, showing the relevance of orthology to evolutionary analysis. It does not discuss distinguishing orthologs from paralogs or characterize that distinction as a fundamental problem. (LLM, conf=0.76)

  cites [6] Jebb D, Six reference-quality genomes reveal evolution of bat adaptations. Nature 583, 578– 584 (2020). 32699395 PMC8075899 sourceExperimental — accuracy not yet measured

In the paper

“Graph-based methods cluster genes into pairs or groups of orthologs based on pairwise sequence similarity such as (reciprocal) best alignment hits (8–12).”

What the sources say

• Supported

  The ‘Identification of Orthologous Groups by OrthoMCL’ section says the method begins with all-against-all BLASTP comparisons and identifies putative ortholog pairs by reciprocal best similarity. It then converts those relationships into a weighted graph and clusters the sequences with MCL. The supplied passages do not give page numbers. (LLM, conf=0.94)

  cites [8] Li L, Stoeckert CJ Jr., Roos DS, OrthoMCL: identification of ortholog groups for eukaryotic genomes. Genome Res 13, 2178– 2189 (2003). 12952885 PMC403725 source
• Supported

  The Introduction says OMA infers orthologous pairs from protein sequences and groups genes into orthologous sets. Materials and methods (§2.2) describes comparing distances to find mutually closest genes, while §2.4 describes an orthology graph whose pairwise links are clustered into groups. The Introduction also identifies bidirectional best hits as an orthology-inference method. (LLM, conf=0.90)

  cites [9] Train CM, Glover NM, Gonnet GH, Altenhoff AM, Dessimoz C, Orthologous Matrix (OMA) algorithm 2.0: more robust to asymmetric evolutionary rates and more scalable hierarchical orthologous group inference. Bioinformatics 33, i75– i82 (2017). 28881964 PMC5870696 source
• Partly supported

  The Algorithm and Software section (page unavailable) says OrthoDB clusters genes into ortholog groups after Smith–Waterman assessments of gene homology. It does not specify graph-based clustering or reciprocal best alignment hits; searches for those terms found no supporting passage in our copy. (LLM, conf=0.85)

  cites [10] Zdobnov EM, OrthoDB v9.1: cataloging evolutionary and functional annotations for animal, fungal, plant, archaeal, bacterial and viral orthologs. Nucleic Acids Res 45, D744– D749 (2017). 27899580 PMC5210582 sourceExperimental — accuracy not yet measured
• Supported

  The Construction of Hierarchical Orthologous Groups section says eggNOG computes pairwise Smith–Waterman similarities and joins proteins into orthologous groups using triangles of reciprocal best hits, then bidirectional best hits. This supports the cited description of similarity-based clustering of orthologs. (LLM, conf=0.92)

  cites [11] Jensen LJ, eggNOG: automated construction and annotation of orthologous groups of genes. Nucleic Acids Res 36, D250– 254 (2008). 17942413 PMC2238944 source
• Supported

  The Background describes methods that use pairwise sequence-similarity scores to identify orthologs or orthogroups. Its Fig. 1 caption gives reciprocal best hits as an example of similarity-based ortholog inference. The paper does not use the label “graph-based” in these passages, but it supports the methodological description attributed to it. (LLM, conf=0.86)

  cites [12] Emms DM, Kelly S, OrthoFinder: phylogenetic orthology inference for comparative genomics. Genome Biol 20, 238 (2019). 31727128 PMC6857279 source

In the paper

“Current methods for orthology inference are either based on graph or gene tree approaches or a combination of both (7).”

What the sources say

• Supported

  Section 2 classifies most orthology-inference methods as graph-based or tree-based. Section 2.3 explains that the approaches often overlap in practice and that independent methods can be combined. The supplied passages do not give page numbers. (LLM, conf=0.93)

  cites [7] Altenhoff AM, Glover NM, Dessimoz C, Inferring Orthology and Paralogy. Methods in molecular biology 1910, 149– 175 (2019). 31278664 source

In the paper

“Gene tree-based methods determine whether the evolutionary lineages of two genes coalesce in a speciation or a duplication node (12–14).”

What the sources say

• Supported

  The Methods section, “Ortholog inference and identification of gene duplication events from gene trees,” says that gene-tree nodes are classified as speciation or duplication events, and that genes diverging at those nodes are orthologs or paralogs, respectively. This supports the citing sentence’s description of gene tree-based methods. (LLM, conf=0.94)

  cites [12] Emms DM, Kelly S, OrthoFinder: phylogenetic orthology inference for comparative genomics. Genome Biol 20, 238 (2019). 31727128 PMC6857279 source
• Supported

  The Introduction describes gene trees as evolutionary histories used to detect duplications and infer orthology and paralogy. The Benchmarking section describes pairwise gene-relationship predictions based on detecting duplications, while New Data and Visualization Features identifies speciation and duplication nodes in those trees. Together, these support the citing sentence’s description of gene tree-based methods. (LLM, conf=0.88)

  cites [13] Huerta-Cepas J, Capella-Gutierrez S, Pryszcz LP, Marcet-Houben M, Gabaldon T, PhylomeDB v4: zooming into the plurality of evolutionary histories of a genome. Nucleic Acids Res 42, D897– 902 (2014). 24275491 PMC3964985 source
• Supported

  In “A robust, computationally efficient pipeline for gene tree generation” (page unavailable), the paper describes reconciling gene trees with a species tree, calling duplications on internal nodes, and deriving pairwise ortholog and paralog relationships. “Web display” (page unavailable) explicitly distinguishes duplication nodes from speciation nodes. Together, these support the claim’s description of how gene tree-based methods classify where two gene lineages meet. (LLM, conf=0.90)

  cites [14] Vilella AJ, EnsemblCompara GeneTrees: Complete, duplication-aware phylogenetic trees in vertebrates. Genome Res 19, 327– 335 (2009). 19029536 PMC2652215 source

In the paper

“Second, we used TOGA with the mouse GENCODE M25 annotation (22) as the reference. This resulted in a higher BUSCO completeness for 98.5% (Ensembl) and 64% (NCBI) of the species (Fig. 3A,B, Tables S6–S7).”

What the sources say

• Supported

  The citation is attached to the identification of mouse GENCODE M25 as the reference annotation, not to the TOGA or BUSCO results. The Data Access section identifies GENCODE M25 as the mouse release; the completeness percentages are presented as the citing paper’s own results. (LLM, conf=0.92)

  cites [22] Frankish A, Gencode 2021. Nucleic Acids Res 49, D916– D923 (2020). PMC7778937 33270111 source

In the paper

“TOGA annotations have a higher completeness score for 97% (Ensembl) and 91.5% (NCBI) of the species (Fig. 3A,B, Tables S6–S7), increasing annotation completeness of conserved genes by an average of 4.1% or ~377 genes (Ensembl) and 0.7% or ~64 genes (NCBI) (fig. S30).”

What the sources say

• ?Not checked — no full text of the sourcecites [20] Thibaud-Nissen F, P8008 The NCBI Eukaryotic Genome Annotation Pipeline. Journal of Animal Science 94, 184– 184 (2016). source
• ?Could not verify

  The Introduction and Vertebrates sections describe RefSeq annotation and curation, but do not report a comparison with TOGA annotations. Searches for TOGA, BUSCO, Ensembl, and the stated percentages and gene counts found no passage addressing the claimed completeness results. (LLM, conf=0.78)

  cites [21] O’Leary NA, Reference sequence (RefSeq) database at NCBI: current status, taxonomic expansion, and functional annotation. Nucleic Acids Res 44, D733– 745 (2016). 26553804 PMC4702849 sourceExperimental — accuracy not yet measured
• ?Could not verify

  The Gene Annotation Updates and Conclusion sections discuss improvements to GENCODE’s human and mouse annotations, not a comparison of TOGA annotations with Ensembl and NCBI across species. Searches for TOGA, BUSCO, and the claimed percentages and gene counts found no passage reporting those results in our copy. (LLM, conf=0.76)

  cites [22] Frankish A, Gencode 2021. Nucleic Acids Res 49, D916– D923 (2020). PMC7778937 33270111 sourceExperimental — accuracy not yet measured
• Partly supported

  The Upgrade of Data Sets and Benchmark of Predictions sections describe BUSCO scores as estimates of genome and gene-set completeness, supporting the use of BUSCO as the assessment method. Our copy does not report a TOGA comparison with Ensembl or NCBI annotations, or the claimed percentages and gene counts. No page numbers are available in our copy. (LLM, conf=0.86)

  cites [23] Manni M, Berkeley MR, Seppey M, Simao FA, Zdobnov EM, BUSCO Update: Novel and Streamlined Workflows along with Broader and Deeper Phylogenetic Coverage for Scoring of Eukaryotic, Prokaryotic, and Viral Genomes. Molecular biology and evolution 38, 4647– 4654 (2021). 34320186 PMC8476166 sourceExperimental — accuracy not yet measured

In the paper

“Using orthologs between human and three representative mammals (rat, cow, elephant), TOGA detected 97.6%, 98.9% and 96.5% of the orthologs provided by Ensembl (Fig. 2A, Table S5), showing a good agreement.”

What the sources say

• Partly supported

  The 'Overlap of ortholog sets' section describes EnsemblCompara ortholog predictions, supporting the identification of Ensembl as an ortholog source. It does not report a TOGA comparison or the 97.6%, 98.9%, and 96.5% detection rates; those results are not supported by this paper. (LLM, conf=0.88)

  cites [14] Vilella AJ, EnsemblCompara GeneTrees: Complete, duplication-aware phylogenetic trees in vertebrates. Genome Res 19, 327– 335 (2009). 19029536 PMC2652215 sourceExperimental — accuracy not yet measured

In the paper

“Orthologous chains tend to cover several genes located in a conserved order, resulting in higher synteny values, which can help to distinguish orthologs from paralogs (14, 41–43).”

What the sources say

• Partly supported

  The Gene synteny metric section (page unavailable) says that ordered, flanking orthologous genes indicate conserved gene order, and the Comparison using the synteny metric section uses that measure to assess prediction specificity. The paper does not describe orthologous chains or show that they have higher synteny values than paralogous chains; it also says synteny was not used directly to build its trees. (LLM, conf=0.82)

  cites [14] Vilella AJ, EnsemblCompara GeneTrees: Complete, duplication-aware phylogenetic trees in vertebrates. Genome Res 19, 327– 335 (2009). 19029536 PMC2652215 sourceExperimental — accuracy not yet measured
• Partly supported

  The Background defines conserved synteny as maintenance of gene content and order, and the Results (Algorithms and Implementation; page unavailable) say gene-order scores help distinguish putative orthologous from paralogous regions. The paper does not establish that orthologous chains generally span several genes or have higher synteny values than paralogous chains. (LLM, conf=0.78)

  cites [41] Lehmann J, Stadler PF, Prohaska SJ, SynBlast: assisting the analysis of conserved synteny information. BMC Bioinformatics 9, 351 (2008). 18721485 PMC2543028 sourceExperimental — accuracy not yet measured
• Partly supported

  The Results (‘Using Local Synteny to Infer Orthology,’ page unavailable) define synteny by matches among neighboring genes, and the Conclusions say local synteny can distinguish true orthologs from recent retrogenes. But the Methods (‘Local Synteny,’ page unavailable) explicitly say those matches need not be co-linear, so the paper does not support conserved gene order as a requirement or describe the claimed orthologous chains. (LLM, conf=0.87)

  cites [42] Jun J, Mandoiu II, Nelson CE, Identification of mammalian orthologs using local synteny. BMC Genomics 10, 630 (2009). 20030836 PMC2807883 sourceExperimental — accuracy not yet measured
• Partly supported

  The Introduction says conserved gene neighborhoods can help distinguish orthologs from out-paralogs, and the Method describes checking multiple neighboring genes, with one routine considering gene order. Our copy does not state that orthologous chains tend to span several genes or that they produce higher synteny values. The indexed passages provide no page numbers. (LLM, conf=0.82)

  cites [43] Jahangiri-Tazehkand S, Wong L, Eslahchi C, OrthoGNC: A Software for Accurate Identification of Orthologs Based on Gene Neighborhood Conservation. Genomics Proteomics Bioinformatics 15, 361– 370 (2017). 29133277 PMC5828658 sourceExperimental — accuracy not yet measured

In the paper

“CESAR has a high accuracy in correctly aligning shifted splice sites, is able to detect precise intron deletions that merge two neighboring exons, and generates alignments of intact exons (defined as exon alignments with consensus splice sites and an intact reading frame) whenever possible (18, 19).”

What the sources say

• Partly supported

  The Introduction reports that CESAR aligned shifted splice sites with 91% accuracy and sought alignments with intact reading frames and consensus splice sites. But it says the original CESAR failed to annotate exons joined by complete intron deletions; the Results attribute precise annotation of those joined exons to CESAR 2.0's new gene mode, not to CESAR generally. (LLM, conf=0.89)

  cites [18] Sharma V, Schwede P, Hiller M, CESAR 2.0 substantially improves speed and accuracy of comparative gene annotation. Bioinformatics 33, 3985– 3987 (2017). 28961744 sourceExperimental — accuracy not yet measured
• Partly supported

  The Results (page not provided) report high accuracy for aligning shifted splice sites, and the Discussion says CESAR balances preserving splice sites and reading frames whenever possible against reporting real mutations. The Results also identify precise intron deletions among mutations remaining after realignment, but attribute that identification to the authors’ investigation; they do not establish that CESAR itself detects deletions that merge neighboring exons. (LLM, conf=0.83)

  cites [19] Sharma V, Elghafari A, Hiller M, Coding exon-structure aware realigner (CESAR) utilizes genome alignments for accurate comparative gene annotation. Nucleic Acids Res 44, e103 (2016). 27016733 PMC4914097 sourceExperimental — accuracy not yet measured

In the paper

“First, by detecting inactivated genes and providing orthologous sequences for codon alignments, TOGA enables phylogenomic analyses as well as screens for selection patterns and gene losses that are linked to relevant phenotypes (6, 35–38).”

What the sources say

• Partly supported

  The Methods > Phylogenomics section says TOGA classified 12,931 genes as orthologues for sequence alignments and phylogenomic analysis; Methods > Gene selection, loss and gain describes a selection screen of those genes. The paper also reports gene-loss screens linked to bat traits, but attributes the loss-screening procedure to a previously developed approach, not explicitly to TOGA. Our copy therefore does not establish the TOGA-enabled gene-loss part of the claim. (LLM, conf=0.82)

  cites [6] Jebb D, Six reference-quality genomes reveal evolution of bat adaptations. Nature 583, 578– 584 (2020). 32699395 PMC8075899 sourceExperimental — accuracy not yet measured
• ?Could not verify

  The Materials and Methods section, “Identifying Sites Undergoing Positive Selection” (page unavailable), describes codon alignments generated with pal2nal and selection analyses performed with CodeML. Searches for TOGA, inactivated genes, and gene losses found no passage attributing these capabilities to TOGA; the claim therefore cannot be verified from our copy. (LLM, conf=0.76)

  cites [35] Damas J, Broad host range of SARS-CoV-2 predicted by comparative and structural analysis of ACE2 in vertebrates. Proceedings of the National Academy of Sciences of the United States of America 117, 22311– 22322 (2020). 32826334 PMC7486773 sourceExperimental — accuracy not yet measured
• Partly supported

  The Gene identification section says TOGA identifies orthologous gene loci, and Screen for gene losses says it detects gene-inactivating mutations in a search for limb-related gene losses. These passages support the orthology and phenotype-linked gene-loss parts of the claim, but do not establish that TOGA provides codon alignments or enables phylogenomic analyses and screens for selection patterns. Page numbers are unavailable in our copy. (LLM, conf=0.85)

  cites [36] Roscito JG, Convergent and lineage-specific genomic differences in limb regulatory elements in limbless reptile lineages. Cell Rep 38, 110280 (2022). 35045302 sourceExperimental — accuracy not yet measured
• Supported

  The Methods sections describe using TOGA to identify orthologous genes and gene losses, and its codon alignments to test for relaxed selection. The Results and Discussion link identified gene losses to blood-feeding adaptations and other vampire-bat traits. Page numbers are not provided in our copy. (LLM, conf=0.91)

  cites [37] Blumer M, Gene losses in the common vampire bat illuminate molecular adaptations to blood feeding. Sci Adv 8, eabm6494 (2022). PMC8956264 35333583 source
• Partly supported

  The Materials and methods sections on investigating SERPINE3 losses and selection analysis say TOGA classifies potentially lost genes and supplies codon sequences used to test selection. The phenotype-linked gene-loss screen used a previously generated dataset and the Forward Genomics approach, rather than TOGA. Thus the paper supports TOGA’s role in gene-loss detection and selection analysis, but not the attributed role in that screen. (LLM, conf=0.83)

  cites [38] Indrischek H, Vision-related convergent gene losses reveal SERPINE3’s unknown role in the eye. eLife 11, (2022). PMC9355568 35727138 sourceExperimental — accuracy not yet measured

In the paper

“CESAR 2.0 is a Hidden Markov model-based method that takes the coding exons of the reference species as input and considers reading frame and splice site information when generating exon alignments in the query sequence.”

What the sources say

• Supported

  The Introduction describes CESAR’s HMM as capturing the reference exon’s reading frame and splice-site annotation to realign a putative exon sequence in the query genome. The Methods state that CESAR 2.0 retains that approach and can align coding-gene exons to a query genomic locus. (LLM, conf=0.91)

  cites [18] Sharma V, Schwede P, Hiller M, CESAR 2.0 substantially improves speed and accuracy of comparative gene annotation. Bioinformatics 33, 3985– 3987 (2017). 28961744 source
• Partly supported

  The Introduction and Materials and Methods (‘Structure of CESAR’; page not available) describe CESAR as an HMM that uses a reference coding exon to realign query sequences while accounting for reading frame and splice sites. The paper describes CESAR, not CESAR 2.0, so it does not establish that the claim applies to that version. (LLM, conf=0.88)

  cites [19] Sharma V, Elghafari A, Hiller M, Coding exon-structure aware realigner (CESAR) utilizes genome alignments for accurate comparative gene annotation. Nucleic Acids Res 44, e103 (2016). 27016733 PMC4914097 sourceExperimental — accuracy not yet measured

In the paper

“Transcript classification is not a straightforward problem, as assembly gaps result in missing parts of the CDS and individual exons can get lost in otherwise clearly conserved genes, as shown in previous work (5).”

What the sources say

• Supported

  The Methods section (page not provided) treats deleted or unaligning exons overlapping assembly gaps as artifacts and states that considered exon deletions occur in an otherwise-conserved context. These passages support both parts of the claim. (LLM, conf=0.90)

  cites [5] Sharma V, A genomics approach reveals insights into the importance of gene losses for mammalian adaptations. Nature communications 9, 1215 (2018). PMC5865188 29572503 source

In the paper

“For the 488 placental mammal and 501 bird assemblies, we provide comparative gene annotations, ortholog sets, lists of inactivated genes and multiple codon alignments generated with MACSE v2 (33) for download at http://genome.senckenberg.de/download/TOGA/.”

What the sources say

• Supported

  The citation is attached to “MACSE v2,” identifying the tool used to generate the codon alignments, not attributing the mammal and bird datasets to the tool's paper. The Brief Communication describes MACSE v2 as a toolkit for producing multiple alignments of protein-coding sequences and shows an alignment displayed at the codon level. (LLM, conf=0.92)

  cites [33] Ranwez V, Douzery EJP, Cambon C, Chantret N, Delsuc F, MACSE v2: Toolkit for the Alignment of Coding Sequences Accounting for Frameshifts and Stop Codons. Molecular biology and evolution 35, 2582– 2584 (2018). 30165589 PMC6188553 source

In the paper

“Given that the basal split of placental mammals is not yet resolved (59)”

What the sources say

• Supported

  Section 3 explicitly says the branching order at the base of Placentalia is unresolved and debated. Section 6 likewise describes the placental root node as a matter of contention among three competing hypotheses. (LLM, conf=0.94)

  cites [59] Foley NM, Springer MS, Teeling EC, Mammal madness: is the mammal tree of life not yet resolved? Philosophical transactions of the Royal Society of London. Series B, Biological sciences 371, 20150140 (2016). PMC4920340 27325836 source

In the paper

“We used the human GENCODE V38 (Ensembl 104) gene annotation (22), which implies that each gene is intact in Boreoeutheria”

What the sources say

• Partly supported

  The Introduction identifies GENCODE as a provider of human gene annotations, and Data Access says its releases are available through Ensembl. But Data Access names GENCODE 35 as the current human release; our copy does not document V38 (Ensembl 104) or say that the annotation implies every gene is intact in Boreoeutheria. (LLM, conf=0.84)

  cites [22] Frankish A, Gencode 2021. Nucleic Acids Res 49, D916– D923 (2020). PMC7778937 33270111 sourceExperimental — accuracy not yet measured

In the paper

“TOGA implements a different paradigm to infer orthologous loci, improves ortholog detection and annotation of conserved genes compared to state-of-the-art methods, and handles even highly-fragmented assemblies.”

In the paper

“Additionally, TOGA detects gene losses, enables selection screens, and automatically provides a superior measure of mammalian genome quality.”

In the paper

“For example, the evolutionary distance between human and other placental mammals and between chicken and other birds is ≤0.55 substitutions per neutral site (fig. S1, Tables S1–S2), explaining why orthologous introns and intergenic regions partially align within these clades (Fig. 1A,E and fig. S2).”

In the paper

“TOGA predicted for the three species 1,532 (rat), 1,711 (cow) and 2,174 (elephant) additional orthologs that are not listed in Ensembl (Fig. 2A).”

In the paper

“Manual inspection showed that misclassified cases include highly-diverged genes, genes that evolved drastic changes in exon-intron structure or protein length, and a lost gene that is compensated by a processed pseudogene copy, which highlights cases of less certain gene conservation (fig. S20–S23).”

In the paper

“We benchmarked the specificity of this approach on 11,161 conserved genes. Only 21, 22, 12 and 21 of these genes are misclassified as inactivated in mouse, rat, cow and dog, respectively, indicating a very high specificity of 99.80–99.89% (Table S4).”

In the paper

“These genes exhibit alignment identity (mean 70%), alignment coverage (mean 83%) and orthology confidence (mean 94%) values that are more similar to orthologs detected by both methods (means 82%, 94%, 99%) (Fig. 2D), supporting that these genes are undetected orthologs.”

In the paper

“For the orthologs detected only by Ensembl, TOGA did identify an orthologous locus in >93% of the cases, but detected either reading frame inactivating mutations, indicating a lost gene, or that more than half of the coding region overlaps assembly gaps in the query (classified as a missing gene) (Fig. 2C and fig. S27–S28).”

In the paper

“Third, further adding TOGA annotations with generated additional references (cow, horse and cat) increases the total number of annotated genes and detects additional lineage-restricted genes (fig. S32–S34).”

In the paper

“In contrast, synteny is the least important feature, likely reflecting our training data sets that we deliberately enriched with translocated orthologs (fig. S7).”

In the paper

“However, BUSCO’s values saturate at ~97% for highly complete assemblies, while TOGA’s completeness values exhibit a larger dynamic range (Fig. 6A,B), providing a better resolution to distinguish highly contiguous from less contiguous assemblies.”

In the paper

“Assemblies of the same species can suffer from different issues, illustrated by the spotted hyena, where the NCBI GCA_008692635.1 assembly has less missing sequence, but a noticeably higher base error rate compared to the DNAzoo assembly (Fig. 6E).”

In the paper

“Fitting generalized linear models shows that the number of annotated orthologs is positively correlated with assembly quality metrics (contig and scaffold N50) and negatively correlated with the evolutionary distance (substitutions per neutral site) and divergence time (millions of years) to human (fig. S38, Table S12).”

In the paper

“We did not use this feature when training the multi-exon model, as it did not increase classification performance further and hampered the detection of partial lineage-specific duplications of multi-exon genes.”

In the paper

“Demonstrating the effectiveness of TOGA’s gene joining procedure, in the highly-fragmented sperm whale assembly the mean coding sequence length after joining fragmented genes is 97% of the length of the orthologous human gene. This is a substantial improvement over the single largest orthologous fragment present in the assembly (mean 59%) (Fig. 4C, Table S10).”

In the paper

“These thresholds correspond to a sensitivity of 0.98 and a precision of 0.99.”

In the paper

“For example, TOGA detects a higher percentage of genes exhibiting inactivating mutations in the Bos gaurus (gaur, 14.2%) compared to the Bos taurus (cow, 4.3%) assembly, indicating that the B. gaurus assembly has an elevated base error rate, whereas both assemblies are indistinguishable in terms of BUSCO completeness (95.8 vs. 95.5%) (Fig. 6D).”

In the paper

“Our UCSC browser mirror at https://genome.senckenberg.de/ provides the TOGA track functionality for 488 placental mammal, 21 non-placental mammal and 501 bird assemblies.”

In the paper

“However, we observed an equally high identity for orthologous genes joined from two, three or even more fragments (Fig. 4B), indicating a high accuracy.”

In the paper

“This is exemplified by two closely related bats: a high-quality Rhinolophus ferrumequinum and a less-contiguous R. sinicus assembly have similar BUSCO (96.4% vs. 96.3% complete genes) but different TOGA completeness values (94.4% vs. 88.2%) (Fig. 6C).”

In the paper

“Interestingly, applying TOGA with human as the reference to 18 marsupial and two monotreme mammals, where neutrally evolving regions are diverged because of the larger evolutionary distance (~0.8 and ~1 substitution per neutral site between human and marsupials/monotremes), still annotates on average 13,397 and 10,238 orthologs (Fig. 5A,B), primarily because gene order is conserved (fig. S43).”

In the paper

“Finally, illustrating extreme cases among seals, 56% of the genes in the Antarctic fur seal have inactivating mutations and 31% of the genes in the Weddell seal have missing exonic sequence (Fig. 6F).”

In the paper

“Similarly, the dog canFam5 assembly exhibits an elevated base error rate compared to dog canFam4 or dingo, whereas all three assemblies have similar BUSCO scores (Fig. 6E).”

In the paper

“The final input annotations that TOGA used with human as the reference comprised 39,664 transcripts of 19,464 genes. For mouse, input annotations comprised 33,460 transcripts of 22,257 genes, and for chicken 38,252 transcripts of 18,039 genes.”

In the paper

“The best-fitting model, determined through model selection according to the Akaike Information Criterion (AIC; Table S12), was a negative binomial GLM that included all four predictors.”

In the paper

“The variance-function-based R2 value (58), which we calculated using the R package rsq (https://CRAN.R-project.org/package=rsq, version 2.2), was 11.2%.”

In the paper

“By varying one predictor at a time and keeping the remaining predictors fixed at their mean values (fig. S38), we found that the most influential variable was contig N50, whereas the least influential was scaffold N50.”

• Also stated in: § Generalized linear models · plain

  “Examining the distribution of AIC values across candidate GLMs (Table S12) led to the same conclusion.”

In the paper

“which provides a BUSCO completeness of 99.5%.”

In the paper

“Performing the same analysis after excluding Hominoidea (Apes) led to qualitatively identical results and only slightly different model coefficients, P-values, and R2 values”

In the paper

“This replacement enables CESAR to align such TGA stop codons to sense or to stop codons.”

In the paper

“This can arise if the real orthologous gene is entirely missing in the assembly (thus only a paralog aligns) or if TOGA misclassifies the orthologous gene because of excessive divergence of intronic/intergenic regions.”

In the paper

“Additionally, TOGA’s classification of ancestral genes provides a useful assembly quality benchmark.”

In the paper

“These large frame-preserving deletions result in substantially shorter but likely functional proteins (though it is not known whether the function is truly conserved).”

In the paper

“however, including problematic transcripts such as fusion transcripts or potential NMD targets can lead to wrong gene classifications or orthology types.”

In the paper

“If the locus represented by the paralogous chain does not receive any annotation via an orthologous chain, then TOGA also annotates a gene at this locus (shown in fig. S6), as this locus likely encodes a gene.”

In the paper

“Second, TOGA can provide an initial annotation of conserved genes for newly-sequenced genomes or be integrated together with available transcriptomics data and ab initio gene predictions to comprehensively annotate conserved and lineage-specific genes.”

In the paper

“Exons not spanned by an orthologous chain are also classified as missing (M), as such cases are often due to assembly fragmentation and incompleteness.”

In the paper

“To evaluate the performance of the models in detecting translocated or inverted orthologous genes, we separately tested them on real orthologous genes (typically high synteny values) and artificially rearranged orthologous genes (typically low synteny values of log10(1)) (Fig. 1F, Table S3).”

In the paper

“Importantly, a gene in the reference can have several transcripts (isoforms) and a given gene can have several inferred orthologous loci in the query.”

In the paper

“since this procedure results in a consistent classification of transcripts that have the same inactivating mutations and only differ in the amount of missing sequence.”

In the paper

“This step reduces redundancy as TOGA only annotates the CDS.”

In the paper

“TOGA implements two features that help to avoid codon alignment errors. First, TOGA masks all gene-inactivating mutations such as frameshifts, which otherwise can result in misalignments (fig. S40). Second, whereas existing methods align entire orthologous coding sequences, TOGA is aware of orthology at the exon level. This enables an “exon-by-exon” procedure that generates alignments by aligning”

In the paper

“Because selection in UTR exons is variable, alignments overlapping UTR exons are ignored for feature computation.”

In the paper

“This allows TOGA to correctly annotate and distinguish nested genes on the same strand and overlapping genes located in antisense orientation (fig. S24B and C).”

In the paper

“We tested the single- and multi-exon model on independent test sets obtained for three representative placental mammals that include both a close sister species to mouse (rat) and more distant outgroups (dog, armadillo).”

In the paper

“To distinguish between classes P, M, and D, TOGA utilizes that an orthologous chain provides not only the orthologous query locus, but the aligning blocks of the chain also provide information about the location of individual exons.”

In the paper

“This step is necessary as the Viterbi algorithm used in CESAR’s HMM may also output alignments of exons that do not exist in the query locus, either because the exon is truly deleted or diverged to an extent that no meaningful alignment is possible (class D) or because the exon overlaps an assembly gap in the query genome (class M).”

In the paper

“Similarly, as exemplified in Fig. 1H, TOGA can detect four orthologous loci in the query, but if only one of these loci encodes a functional gene, this results in a 1:1 orthology relationship.”

In the paper

“Nevertheless, for these more distant clades, human is not a powerful reference and a marsupial and a monotreme mammal should be used as the reference instead.”

In the paper

“For orthology inference, individual predicted transcripts need to be consolidated into predicted genes.”

In the paper

“Most of these features quantify how well intronic and intergenic regions, which largely evolve neutrally, align in comparison to coding exons, which largely evolve under purifying selection.”

In the paper

“Since knowledge about U12 introns in the reference may be incomplete or not always available, TOGA considers every intron in the reference without canonical GT/GC-AG splice sites as a putative U12 intron.”

In the paper

“We used the NCBI RefSeq identifier and the assembly name provided by Ensembl to assure that all comparisons between TOGA and NCBI or Ensembl were done for the same genome assembly.”

In the paper

“Since TOGA but not Ensembl distinguishes between 1:many (more than 1 ortholog in the query species) and many:1 (one ortholog in the query, but more than one in the reference), we updated those Ensembl 1:many types as many:1, for which the orthology group had >1 gene annotated in reference and exactly one gene annotated in the query.”