Changelog

Version 0.6.1 expands clinical interpretation support with richer secondary-finding validation, stronger HPO extraction, genome-build safeguards, and clearer reports.

• A new "Analyze the description by AI to extract the HPO terms" button now sits under the clinical description on the Rare disease, Carrier screening, and Trio analysis forms. One click asks the AI to extract present and explicitly absent HPO terms, fills the dedicated included/excluded fields, and shows every suggestion as a removable chip so users can drop anything incorrect before running.
• HPO extraction is more sensitive and extensible: clinical paraphrases are recognized, related concepts are surfaced (for example migraine also yields its direct parent Headache), and AI output must use valid identifiers from the loaded ontology. Weak or co-occurring features such as visual impairment, photophobia, or nausea are only marked present when the source text actually supports them.
• Phenotype-to-gene matching now uses the HPO hierarchy: a specific patient term can support a broader associated gene annotation (migraine keeps genes annotated with headache in scope).
• AI-assisted Word report writing now covers Carrier screening and Trio analysis as well as Rare disease, producing narrative, descriptive DOCX reports with a labeled AI-Assisted Narrative section and a deterministic fallback when AI is disabled or unavailable.
• During Step 4 (AI variant review) a second live progress indicator shows the analyzed/total candidate count beneath the overall progress bar.
• The LitVar literature table in the Rare disease pipeline no longer shows the Expression column (kept in Trio), and report rendering and layout were polished across pipelines.
• ACMG secondary-finding review now uses the curated SF v3.3 gene–disease rules, including gene-specific inheritance and variant-count requirements before a finding is reported.
• Genome-build guard rails check uploaded metadata, contig evidence, coordinates, and—when appropriate—variant overlap with the selected BED regions to flag suspicious hg19/hg38 mismatches before analysis.
• Clinical descriptions can be analyzed with AI to extract ontology-validated included and excluded HPO terms, while the HPO matcher recognizes clinically related concepts without automatically adding weakly supported findings.
• Report presentation was refreshed with clearer result panels, improved phenotype chips, theme-aware logs, and Inter typography across the web application.

Version 0.6.0 adds an on-demand AI phenotype extractor to all three pipelines, extends AI-assisted narrative Word reports to Carrier screening and Trio analysis, and adds a live AI review progress indicator while analysis runs.

• A new "Analyze the description by AI to extract the HPO terms" button now sits under the clinical description on the Rare disease, Carrier screening, and Trio analysis forms. One click asks the AI to extract present and explicitly absent HPO terms, fills the dedicated included/excluded fields, and shows every suggestion as a removable chip so users can drop anything incorrect before running.
• HPO extraction is more sensitive and extensible: clinical paraphrases are recognized, related concepts are surfaced (for example migraine also yields its direct parent Headache), and AI output must use valid identifiers from the loaded ontology. Weak or co-occurring features such as visual impairment, photophobia, or nausea are only marked present when the source text actually supports them.
• Phenotype-to-gene matching now uses the HPO hierarchy: a specific patient term can support a broader associated gene annotation (migraine keeps genes annotated with headache in scope).
• AI-assisted Word report writing now covers Carrier screening and Trio analysis as well as Rare disease, producing narrative, descriptive DOCX reports with a labeled AI-Assisted Narrative section and a deterministic fallback when AI is disabled or unavailable.
• During Step 4 (AI variant review) a second live progress indicator shows the analyzed/total candidate count beneath the overall progress bar.
• The LitVar literature table in the Rare disease pipeline no longer shows the Expression column (kept in Trio), and report rendering and layout were polished across pipelines.

Version 0.5.0 adds a per-gene organ-expression section to the ranked variants table, powered by a new VarSage REST endpoint backed by the GTEx Portal API v2 (dataset gtex_v8, GENCODE v26).

• Every gene in the Rare disease, Carrier screening, and Trio analysis result tables now has an Expression chip; opening it fetches the same GTEx evidence and expands an inline panel.
• The new /api/expression REST endpoint resolves the symbol against the GTEx reference-gene set, then requests per-tissue median TPM from the GTEx clusteredMedianGeneExpression endpoint, returning 53 tissue sites with UBERON ontology terms.
• The panel renders a horizontal per-tissue bar chart ordered from the highest- to lowest-expressing organ, with a human-readable tissue name and TPM value on every row.
• Each panel also includes an animated grow-in effect, an auto-written insight sentence (median, peak tissue, tissues above 1 TPM), an editable reviewer insight field, and a deep link into the GTEx portal.
• The previous ENCODE expression source has been fully removed; GTEx is now the single, more reliable source of tissue-level expression evidence.

Version 0.4.5 makes phenotype extraction more tolerant of clinically equivalent wording while keeping the extraction evidence-bounded and ontology-validated.

• Deterministic HPO text matching now recognizes a small curated set of clinical paraphrases, informal phrases and common wording variants, such as developmental delay/psychomotor delay, consanguineous parents/first cousins, neurodevelopmental problems, and mucolipidosis type II/III.
• Theoretical basis: natural clinical language is many-to-one with ontology concepts, so exact HPO label matching alone can lose true phenotype evidence. Curated aliases expand lexical recall without changing the underlying HPO concept or creating a new phenotype.
• AI extraction remains constrained by ontology membership and requires evidenceText supported by the supplied patient text. The extractor continues to reject unsupported IDs, disease or gene names used as evidence, and inferred absent findings.
• Whole-phrase and word-boundary matching, plural normalization, explicit-negation handling, and excluded-finding precedence remain in place to limit false positives.

Rare-disease analyses now end with a cautious, deterministic case-level decision (positive candidate finding, inconclusive candidate finding, or no positive candidate identified) shown on the results page and in the HTML and Word reports, explained with an evidence spectrum and a reading guide. This release also includes account-email improvements, per-pipeline DP/GQ filter thresholds, and BED region filtering across all pipelines.

• Rare-pipeline reports now carry a cautious case-level decision derived from ACMG/reportability and ClinVar evidence, with careful wording that keeps the result as decision support pending expert review and confirmation.
• The results page and HTML report present the decision on a green-to-red evidence spectrum with a marker and status-colored panel, plus a built-in guide explaining each outcome; Word reports now include the decision as well.
• Read depth (DP) and genotype quality (GQ) thresholds can now be set by users on all three pipeline submission forms (rare disease, carrier screening, trio).
• BED region filtering is available on all three pipelines with the Twist exome capture panel preselected by default.

Gene inheritance modes are now read completely and correctly from the HPO database, and when a gene has several modes the patient's phenotype helps select the right one.

• Fixed the HPO term mapping: semidominant inheritance now uses its correct term, and uniparental disomy is no longer mislabeled.
• Added previously missing inheritance modes (Y-linked, typically de novo, somatic mosaicism, polygenic, digenic, and others) so they are no longer miscounted as phenotypes during matching.
• De novo is now a first-class inheritance mode in scoring, and equal-score ties between a gene's modes resolve deterministically.
• When a gene has several inheritance modes, the patient's phenotype now selects the mode whose disease best fits the case (for example SRY's Y-linked versus X-linked dominant modes).
• Disease-level inheritance terms from HGNC-linked OMIM/ORPHA diseases are now merged into gene profiles.

Gene names in analysis tables now open OMIM directly, and the website now shows a changelog page plus a dismissible update banner when new releases land.

• Gene cells in ranked, carrier, trio, and report tables now link to OMIM.
• The new changelog page at /changelog lists recent website updates.
• Returning users see a small announcement banner after updates until they dismiss it once.

Recent workflow refinements focused on report readability, review guidance, and safer interpretation support.

• Improved report wording and review summaries.
• Expanded help text for analysis and export flows.
• Tighter layout behavior across result tables and cards.