Source: https://docs.nygen.io/CyteType/results/report/ Revision: 6ce97c558d4353ff1796056a86b1e6d8b42f4b4e Structured page: https://docs.nygen.io/CyteType/results.report.json The CyteType report Every job has an interactive report at https://cytetype.nygen.io/report/. It shows your embedding, the label for each cluster and the evidence behind it: the candidates CyteType considered, the marker genes, the reviewer’s assessment and how the label fits the study. The labels are also copied into your object (see Results in AnnData and Seurat ); the report is where you check them. Open the report The client prints the report link as soon as the job starts, so you can follow the job before it finishes. The link is also stored with the results. Python R adata.uns["cytetype_jobDetails"]["report_url"] To open a report by its job ID from a terminal: cytetype view obj@misc$cytetype_jobDetails$report_url ViewCyteTypeJob("") If you installed the command-line launcher with InstallCyteTypeRCli(), cytetyper view does the same from a terminal. cytetype view and ViewCyteTypeJob() open the report through the sign-in page of the server you signed in to; if your browser is already signed in, you go straight to the report (see the command-line reference ). You can also open any of your reports from the Reports list in the dashboard with View. Who can open a report The report uses your browser sign-in (the emailed one-time code at https://cytetype.nygen.io/login), not your API key. Who can open it depends on the job’s visibility: Visibility Who can open the report Private (the default) You, and the people you shared it with by email. Organization Members of your organization’s license, and the people you shared it with. Public Anyone with the link, without signing in. Any signed-in user can also find it in Chat’s report search. If you are not signed in, the report shows Authentication Required with a Sign In to View Report button, and after signing in you return to the report. An account without access sees the same screen even when it is signed in: ask the owner to share the report with that email address. A deleted report shows Report Deleted, and a job ID that does not exist shows Error Loading Report. Visibility and sharing explains how to change the visibility and manage shares, and what a public report exposes. Follow a running job The report opens while the job is still running and updates itself, so you don’t need to reload the page. With no cluster selected, Study Information shows a progress tracker with the job’s steps: Preparing, Understanding context, Annotating clusters, Refining annotations, Reviewing and Finalizing. During Annotating clusters and Reviewing it also counts how many clusters have passed that step. In the cluster list, a spinner marks each cluster while the job runs (a clock before it starts). Selecting one shows Annotation Pending or Annotation in Progress with its current step, for example Annotating or Refining. The embedding is available from the start. The labels and the evidence appear when the job reaches its final step. Gene search becomes available once the report’s gene index is ready; until then the search box reads “Loading gene index…”. Page layout The report has three columns: Left sidebar: the cluster list, with sorting, filters, a label switch and search. Center: gene search, the embedding, and a card for the selected cluster (or the study information). Right: a toolbar (Share, Open Chat, Download as CSV) above the evidence for the selected cluster, or the study summary when no cluster is selected. Select a cluster in the list or click it on the embedding. The address bar then includes ?cluster=, so you can bookmark or send a link to a specific cluster, and the browser’s back and forward buttons move between the clusters you viewed. The up and down arrow keys move to the previous or next cluster in the list. Note Reports of jobs run with earlier CyteType versions can show a tabbed panel (Markers, Justification, Review, Clinical Info, Cluster Context) instead of the sections described on this page. Left sidebar The toolbar at the top of the sidebar has four controls: Control Options Sort Biological Order groups clusters under display-group headers. Alphabetical sorts by label. Cluster ID sorts by CyteType’s cluster ID. Filter Confidence: High, Moderate, Low. Match: High (≥80%), Moderate (40-79%), Low (<40%), No Original Label. Labels CyteType Label or User Label (the labels in your cluster column). Search clusters... Filters the list by cluster ID or label as you type. Esc clears it. Each row shows CyteType’s cluster ID, the label and three dots. The dots show the reviewer’s confidence, or the Match score when you display user labels; hover over them to see the value. Clusters that are still running are always listed. Clusters without a confidence value are hidden as soon as you untick any of the confidence boxes. In Biological Order, each header names a display group and counts its clusters. CyteType picks the level of its reference hierarchy for these groups so that the study has a manageable number of them, puts unresolved clusters in a group of their own, and lists each identity outside the hierarchy under its own label. The headers appear once the job has finished. While a cluster is selected, hover over a header to highlight its group on the embedding. Dataset Overview, below the toolbar, deselects the cluster and shows the study summary . The footer links to your reports in the Dashboard, the Citation (the CyteType preprint), a light and dark theme switch, and the CyteType repository on GitHub. Center column Gene search Type a gene symbol in the search box (“Search gene expression”), then pick a suggestion or press Enter. The embedding colors each cluster by the percentage of its cells that express the gene, on a scale from 0% to 100%; hover over a cluster to see its value in the legend. Only the genes in your data can be found. Click the gene name in the legend for a short description of the gene from MyGene.info (the top match, not filtered by species), and click × to clear the coloring. Gene names in the right panel work the same way: click a gene to color the embedding by it, and click it again to clear. Embedding The embedding shows the 2D coordinates sent with the job (UMAP by default) for a sample of up to 1,000 cells per cluster (max_cells_per_group), and at most 50,000 points in all. This sampling affects only the display. The embedding cannot be zoomed or panned. Hover over a cluster to see its label, and click it to select it. When cluster shapes are available, a button below the embedding cycles the view between Scatter (cells only), Hybrid (cells over the cluster shapes) and Hull (shapes only), and another one shows abbreviated cluster labels, which you can drag into place; Reset label positions puts them back. The shapes are created at the end of the job, so reload a report that you opened while the job was running. If the job had no coordinates, the embedding area reads “UMAP Visualization Not Available”. Cluster card When a cluster is selected, a card below the embedding summarizes it. Click the title, the subtitle, the cell state or the confidence to jump to the matching section of the right panel. Field Content Title The CyteType label. Opens Justification. Subtitle The granular annotation, a one-sentence description of the cluster. Opens Cell Type Evidence. Cell State The cell state, if one was called. Opens Cell State Evidence. Confidence The reviewer’s confidence, with a Heterogeneous badge when the reviewer found distinct populations in the cluster. Opens Quality Control. Ontology The Cell Ontology ID and term name, linked to the term in the EBI Ontology Lookup Service. User Label Your label for the cluster, with its Match score when there is one. Study information With no cluster selected, the card shows Study Information: the job Status (with the progress tracker while the job runs), a Completed row with the time the job was submitted, the User-Provided Metadata you passed as metadata (values that are URLs become links), and the Client that submitted the job with its version, for example anndata v0.21.0 or seurat v0.6.0. Study context and metadata explains which metadata to pass. Toolbar The toolbar sits above the right panel. Button What it does Share Shown only to the report’s owner. Invites one person by email; they get view access. Manage all shares in dashboard opens your reports in the dashboard. Open Chat Opens Chat in a new tab with this run attached, and with the selected cluster if there is one. Chat needs you to be signed in; the button reads Sign in required when you are not. Download as CSV Downloads the cluster table described in Download as CSV . People you share a report with by email can open it after they sign in with that address, including people who don’t have an account yet. See Visibility and sharing . Right panel: the evidence for a cluster Select a cluster to see its evidence in up to five sections: Justification, Cell Type Evidence, Cell State Evidence, Quality Control and Context Fit. A navigation bar at the top jumps to each section and adds a small summary: the number of genes for Cell Type Evidence, the number of gene sets for Cell State Evidence, and a colored dot for the confidence in Quality Control. Sections, and most items inside them, collapse and expand when you click their header. Gene chips are colored by how often the gene is detected in this cluster compared with the average of the other clusters: red when it is detected more often, blue when less often. Hover over a chip to see the log2 ratio and the percentage of the cluster’s cells that express the gene. Justification How CyteType arrived at the label. Classification chip. The header shows Unresolved when no identity could be assigned. Reports of older jobs can also show Broader parent, Mixed / doublet or Malignant / aberrant. Interpret and review results explains each. Annotation hierarchy. The label’s place in CyteType’s reference hierarchy of cell types, as a path that starts at L1 (broadest). For an identity outside the hierarchy it shows only the identity. When the label combines two identities, each has its own row. Summary. The conclusion for the label. Python returns the same text as justification. Candidates Investigated. The identities that were considered and rejected, marked with ×. Each shows Evidence against and Why considered, then Evidence details and Miscellaneous. Selection Rounds. The identity selected in each round, marked with a Round chip and ✓. CyteType evaluates candidates in rounds and moves to more specific identities while the evidence supports them. Each round shows Evidence supporting, then either How the annotation was refined (a more specific identity went on to the next round) or Why this level was retained (why the annotation stopped at this level), followed by the details. Label Refinement. When several clusters share an identity, CyteType compares them and can add a qualifier that tells them apart. This item shows the Distinguishing feature and the Reasoning; the CSV export has the qualifier in its own column. Biological order group. For an identity outside the hierarchy, the broader biological group it was placed in and why. The Evidence details of a candidate can include: Item Content Evidence supporting, Evidence against, Why considered The arguments for and against the identity, and why it was proposed. Evidence synthesis A balanced assessment of the supporting, missing and unexpected genes. Pathways Pathways that support the identity. Gene evidence Genes listed as Evidence supporting, Expected but weak or absent and Unexpected expression. Reference The Reference hierarchy path and the Reference match reasoning for placing the identity in the hierarchy. Population consistency Whether the evidence points to one population or to two distinct identities. Miscellaneous holds Review confidence and Reference match confidence percentages, the Reference node identifier and the Reference ranking rationale. A candidate with a combined identity chip merges two identities, for example when the cluster appears to contain both; its Merged identity item lists them. Important The percentages under Miscellaneous rank candidates inside the annotation step. They are not the reviewer’s confidence shown in Quality Control, and they are not probabilities that the label is correct. Reports of older jobs show Other candidates investigated instead of the rounds, with ×××, ×× or × for a strong, moderate or weak rejection, a Consideration reason and a Rejection reason, and the Initial candidates under Supplementary information. Cell Type Evidence The genes behind the label, and how the cluster differs from the clusters closest to it. Overview graphic: the number of Supporting, Missing and Unexpected genes, drawn as a bar. Supporting: genes whose expression supports the label, with an explanation. Missing: genes expected for the label that are weak or absent in the cluster. Unexpected: genes expressed in the cluster that are inconsistent with the label. Differentially Expressed vs Nearest Clusters: one item per nearby cluster, with a volcano plot (log2FC against -log10 pAdj) and a table of the genes with pAdj below 0.05 and an absolute log2FC above 1 (columns Gene, log2FC and pAdj). Hover over a gene in the table to find it in the plot. A chip names the method: Wilcoxon: a cell-level test that treats every cell as independent, so p-values can overstate significance. DESeq2: a pseudobulk test in which samples, not cells, are the replicates. Log2FC: the fold change of the percentage of cells expressing each gene. No test is run, so there is no volcano plot and the table has no p-values. Co-expression Evidence: for pairs of genes, the percentage of cells that express one gene only, both genes or neither, with the phi coefficient (φ, from -1 for mutually exclusive to +1 for always co-expressed) and a chi-squared p-value. It appears, for up to three pairs of a supporting and an unexpected gene, when the reviewer found the cluster heterogeneous and cell-level data was uploaded with the job. Cell State Evidence The cell state, kept separate from the cell’s identity. Chips give the state (State:) and the number of gene sets. Summary explains the state, and Supporting gene sets lists named gene programs with their genes and reasoning. The section is absent when no state was called. Quality Control The independent reviewer’s assessment of the final label. Chips show Confidence (High, Moderate or Low) and Heterogeneity (Heterogeneous or Homogeneous). Item Content Confidence The reviewer’s reasoning for the confidence level. “Confidence details not available.” means the review did not run, or failed, for this cluster (see Reviewer confidence ). Heterogeneity The reasoning for the heterogeneity call. Cell classifier Results of a classifier trained on the marker genes of all clusters and tested on cells it did not see during training. The accuracy chip shows how well the genes separate this cluster’s cells from the others (green from 85%, amber from 65%, red below). The Entropy chip is the mean uncertainty of the per-cell predictions; lower means the cells look more alike. The bars show the labels the classifier gave to this cluster’s cells. It needs cell-level data uploaded with the job. Reviewer Panel Several reviewer perspectives, each with a specialty (for example a T cell immunologist), a one-line conclusion, and a Strength, Weakness, Biological interpretation and Alternatives. These assessments are generated by CyteType’s models, not written by human experts. Markers from External Databases How the cluster’s genes compare with reference marker lists, and the Database markers from SingleCellBase, CellGuide Canonical and CellGuide Computational. Context Fit How the label fits the study and the cluster. Tissue: how the identity fits the tissue and the study context. Cluster: how it fits this cluster’s own context. Cluster metadata has a Summary, Keywords, and the Full Metadata Breakdown (%): the percentage of the cluster’s cells in each category of your categorical cell metadata columns. Pathways Enriched/Deleted: the Supporting pathways for the identity with an explanation and a Pathway summary, followed by Detailed pathway data for each pathway library. Each table lists the Pathway and its normalized enrichment score (NES). The Single sample tab ranks genes by the share of this cluster’s cells that express them; the Differential tab ranks them against the average of all clusters. A positive NES means the pathway’s genes rank high, a negative NES that they rank low. Only pathways with a false discovery rate below 0.01 are listed, up to 50 per library and tab. Study summary With no cluster selected (click Dataset Overview), the right panel shows: Study Overview: CyteType’s description of the study, built from your study context and the data. Python returns the same text as studyContext from get_results(). Marker Gene Expression: a dot plot with one column per cluster, labeled with its cell type and ordered by display group. The rows are up to three key supporting genes per cluster, grouped by display group. Dot size is the percentage of cells that express the gene; color is the mean expression when cell-level data was uploaded with the job, and otherwise the percentage. Hover over a dot to see the values. The plot appears once the job has finished. The embedding colors every cluster in this view, and the center column shows the study information. Compare with your own labels The report treats the labels in your cluster column (group_key) as your labels. Switch Labels to User Label to show them in the cluster list and on the embedding; the dots then show each cluster’s Match score, and the cluster card shows the score next to your label. The Match score is the similarity between your label and CyteType’s label, from 0% to 100%. The filter bands are High (80% or more), Moderate (40% to 79%) and Low (below 40%). Use the Match filter to focus on the clusters where you and CyteType disagree. There is no score when your labels are cluster numbers or generic names (such as 0, C1, cluster 3 or unknown); such clusters count as No Original Label. To compare CyteType with an existing annotation, run it with that annotation column as group_key, with marker genes computed for the same column (see Prepare your data ). Interpret and review results explains how the score is calculated and how to read it. Download as CSV Download as CSV saves cytetype_annotations_.csv, with one row per finished cluster. Clusters that are still running are left out. Column Content cluster_id CyteType’s cluster ID. cell_type The CyteType label. coarse_cell_type The display group (the Biological Order header). cell_state The cell state. cl_term The Cell Ontology term name. cl_term_id The Cell Ontology ID. confidence The reviewer’s confidence: High, Moderate or Low (Unknown when there is none). is_heterogeneous TRUE or FALSE. supporting_genes The key supporting genes, separated by ;. cell_type_abbrv_label The abbreviated label shown on the embedding. color_hex The cluster’s color in the report. author_label Your label for the cluster. author_label_similarity_score The Match score from 0 to 100; empty when there is none. l1, l2, l3, l4 The annotation hierarchy levels. For an identity outside the hierarchy, l1 holds the identity and l2 to l4 are empty. qualifier The distinguishing feature added by label refinement, if any. The CSV leaves out the evidence text, the reviewer panel, the heterogeneity type and level L5. The Python client returns these (see Results in AnnData and Seurat ). See also Interpret and review results : how to read the labels, confidence and heterogeneity. Results in AnnData and Seurat : the same results in your object. Visibility and sharing , Dashboard and Chat .