Changelog

Log transform for LR analysis

• Fix Log transform for LR analysis

  Fixed log transformation to normalised counts.

LR analysis and UMAP controls

• Improved LR analysis: rank-aggregate

  Rank-aggregate support added to logistic regression for improved marker ranking.

• Fix LR results patch

  Clearer analysis params and timestamps on results; fixed multiple runs for one primary grouping with varied secondary group selection.

• Improved UMAP parameter UI

  Usability improvements to UMAP parameter controls on the Analyze page.

UMAP parameter exposure

• Improved UMAP parameter exposure

  Customise UMAP with min_dist and spread on the Analyze page; values appear in analysis info.

CyteType overhaul

• Improved CyteType overhaul

  Secure report links, better job tracking, deduped per-cell annotations on rerun, and a manual retry when annotation import fails.

CyteType integration

• New CyteType integration

  AI-powered cell type annotation via CyteType, with study context and report workflow in Explore.

• New Auto-annotation model upgrade

  Improved automatic cell type detection accuracy and coverage.

Scatter plot legend

• Fix Scatter plot legend

  Legend correctly excludes cells hidden by category filters when viewing another category.

UMAP scatterplot legend fix

• Fix UMAP scatterplot legend fix

  Cluster cell counts in the legend no longer include excluded or hidden cells.

Cluster marker p-values

• Improved Cluster marker p-values

  Markers table now includes per-cluster p-values.

Comparison validation

• Improved Comparison validation

  Faster, more responsive cell-overlap checks when building comparisons.

• Fix Pseudobulking secondary group

  Fixed secondary group option when set to "None".

Violin plots and analysis thumbnails

• Fix Violin plot categories

  Fixed wrong categories in violin plots and related plot rendering.

• Fix Missing thumbnails

  Fixed missing analysis thumbnails.

Report module reference

Clearer docs for ontology mapping, evidence trails, and review workflows.

Soft-delete jobs

• New Soft-delete jobs

  Owners can remove jobs from the workspace without leaving orphaned report links active; connected progress streams receive a deletion event so live views do not keep polling a removed analysis.

Grant access for private API jobs

• New Grant access for private API jobs

  External backends can authorize a signed-in user to view a private report via POST /report/{job_id}/grant_access without putting API tokens in the browser URL. Built for embedding CyteType reports inside another app's auth boundary.

Job visibility at submit

• New Job visibility at submit

  Annotation requests can set report visibility (private / org / public) at job creation, so sharing policy is decided before the run finishes rather than only from the dashboard afterward.

Annotator comparison and MCP tools

• Improved Annotator candidate comparison

  Competing cell-type hypotheses are scored in an arena-style comparison (including subtype alternatives), with clearer gene-set evidence when markers are missing from the measurement panel.

• Improved MCP annotation tools

  Claude and other MCP clients can pull quality-control summaries, expert reviewer notes, and neighbour differential genes for connected jobs, not only coarse cluster lists.

MCP server and Target Discovery access

• New MCP server

  Explore finished analyses from Claude Desktop and other MCP clients at https://mcp.cytetype.nygen.io. Connect up to 10 jobs, query metadata with SQL, look up expression, run differential expression, and read LIANA cell-cell interactions under the same OTP login used by the web app.

• Improved Target Discovery access control

  The Target Discovery tab and trigger endpoints are gated by license tier and job ownership, so drug/target pipelines only run for authorized accounts on jobs they can access.

Target Discovery

• New Target Discovery

  Post-annotation pipeline that builds a per-cluster therapeutic picture from disease context, Open Targets evidence, signaling, and drug landscape, covering what is actionable, what is already pursued, and where novel opportunities remain. Scoped to the study's disease and tissue metadata.

Cell-cell communication and live progress

• New Cell-cell communication (LIANA)

  Rank-aggregate ligand-receptor analysis runs on uploaded expression data (human and mouse) and feeds neighbourhood refinement, so annotation can use communication axes between clusters rather than markers alone.

• New Live job progress stream

  Report UI follows annotation stage and per-cluster status over a server-sent progress stream, so long runs show where the pipeline is instead of a static spinner.

• Improved Dashboard job naming

  Rename jobs inline and copy report IDs from the job list for handoff into notebooks, LIMS notes, or share links.

Research Agent sessions

• Improved Research Agent sessions

  Chat is rebuilt as named research sessions you can create, revisit, and delete; conversation state and chart outputs persist across turns for multi-step questions on the same report.

Differential expression and report plots

• New Differential expression in reports

  Per-cluster DE with volcano plots and lazy-loaded tables (log2 fold change and adjusted p-values), including strategy labels so you can see whether Wilcoxon or pseudobulk logic was used for that comparison.

• New Summary dotplots

  Pre-computed marker-gene dotplots in the report for cross-cluster expression patterns without exporting to Scanpy/Seurat first.

• Improved Email OTP login

  Passwordless login now uses a one-time email code instead of magic links, with safer redirect handling after verification.

Gene expression on the UMAP and export

• New Gene expression on the UMAP

  Type a gene (or click a gene chip) to colour the UMAP by expression, with hull overlays and cluster abbreviations so spatial structure and marker support are visible together.

• Improved Richer CSV export

  Downloads now include cell type, coarse lineage, cell state, heterogeneity flag, supporting genes, abbreviation label, and cluster colour hex for direct use in Seurat/Scanpy metadata or figure legends.

• Improved Co-expression rationale and classifier markers

  Reports surface why gene pairs were selected for co-expression and how many markers entered the classifier check, making the evidence trail easier to audit.

• Improved Unmeasured genes labelled clearly

  Genes absent from the assay panel are marked NOT_IN_PANEL rather than treated as missing data, so annotation does not over-interpret genes that were never measured.