Overview Data & Annotations Search & Explore Batch Annotation AI Agent Release notes

Release notes

A single log of every user-visible change across FAVOR. Each release covers the web portal, the annotation database, the CLI, the research agent, the batch pipeline, search, and the public API. Newest first.

2026.06.02

Minor release

June 2, 2026

Knowledge graph refresh

The platform moved to the rebuilt knowledge graph. New node and relation types landed in the data layer, new gene-page views show them, variant-trait associations were consolidated, and detail-page sections that no longer have backing data were removed.

Platform

NewGene. New Functional Partners view under Gene Annotation puts gene-to-gene relations in one sortable, filterable table: co-expression (CoXPresdb z-score), co-essentiality, genetic interactions, TF regulation, and protein-protein interactions.
NewGene. Expression view adds single-cell expression by cell type (CellxGene Census) next to bulk tissue expression. Sort by specificity, mean expression, or fraction of cells expressing, and group by organ system.
NewGene. Protein Structure view now resolves the gene's protein product: UniProt identity, isoform transcripts (Ensembl), and protein-complex memberships (EBI Complex Portal), each with an outbound link.
NewGene. Disease Portfolio adds a Pharmacogenomic Guidelines table and a variant-centric prioritized targets table (genes prioritized for a disease or phenotype).
RemovedDetail pages. Disease pages drop the Studies tab. Phenotype pages drop the Variants and Studies tabs. The graph relations behind them were retired.
FixedGene. Pathway-leverage map is populated again. Parent and child pathways and member genes were coming back empty after the schema change.
FixedGraph Explorer. Edge sort and per-edge filters work again. The schema response shape changed and the controls were reading stale field names.

Data

New New node types: Protein, Transcript, Complex, Motif, Dataset, Scoreset, and CellLine. They appear in graph traversals now; dedicated pages come later.
New Gene-to-gene relations: co-expression (CoXPresdb v8), co-essentiality, genetic interactions, and TF regulation of target genes (TRRUST v2).
New Single-cell gene expression by cell type, from CellxGene Census 2024.
New Allele-specific disruption of TF binding by variants, from AdAstra.
New Drug-to-drug structural similarity, from ChEMBL fingerprints (Tanimoto).
New Variant-centric gene prioritization for diseases and phenotypes, plus pharmacogenomic guidelines.
New Gene links to its protein, transcripts, complexes, perturbation datasets, and DMS scoresets.
Updated Variant-trait associations were consolidated into a single relation (they used to be split into disease, phenotype, and EFO-trait buckets). Gene-disease scoring now keys on the Open Targets composite score (ot_score), with validity tier, ClinGen and GenCC classifications, and per-source evidence subscores.
Removed Retired the GWAS-Study and EFO-trait (Entity) node types, along with the study-investigates-trait, variant-reported-in-study, and per-bucket gene and variant trait relations.

2026.05.07

Minor release

May 7, 2026

Perturbation and MaveDB pages

Every gene has a Perturbation tab under Gene Annotation — CRISPR essentiality screens and Perturb-seq downstream/upstream effects from the EBI Perturbation Catalogue, with a KPI strip up top so you see what's there before scrolling. Multiplexed assay of variant effect (MAVE) scores have moved to their own page with the full distribution plotted against the active calibration's LOF / intermediate / functional thresholds.

KRAS perturbation tab — KPI strip with 1 Perturb-seq dataset, 6 downstream targets, 1,191 CRISPR screens, essential in 618 lines; downstream-effects table with log2 FC, direction, magnitude, adjusted p-value, cell type, and dataset. — Gene → Gene Annotation → Perturbation for KRAS.

BRCA1 MaveDB scoreset urn:mavedb:00001222-b-2 — 2,271 variants across 5 calibrations, score distribution histogram with LOF, intermediate, and functional regions overlaid from the active calibration. — Gene → Gene Annotation → Variant Effect Maps → scoreset for BRCA1.

Platform

NewGene Annotation. Perturbation tab opens with a KPI strip — Perturb-seq datasets, downstream targets, CRISPR screens, essential-in cell-line count — so you see what's there before scrolling. Data-source caption links out to EBI Perturbation Catalogue and BioGRID ORCS.
NewCRISPR. CRISPR section pulls from BioGRID ORCS — every screen labelled with its perturbation type (CRISPRn knockout, CRISPRi suppression, CRISPRa activation), score type (FDR, Rho, Gamma, gene dependency probability, …) so the number is interpretable, per-row score interpretation on hover, and cell line + tissue + disease in one column. Server-driven filter bar with perturbation-type pills, tissue picker populated from a live facet, and a significant-only toggle — each change refetches against the full 21M-row catalogue rather than narrowing a stale slice.
NewPerturb-seq. Downstream and upstream tables for every gene. Self-perturbations (gene → gene, the experimental positive control) are excluded from the upstream view. Citation hover on the Dataset cell shows study title and year.
NewMaveDB. Multiplexed assay of variant effect scores live on their own page (Gene Annotation → Variant Effect Maps), pulled straight from MaveDB rather than via the perturbation catalogue. Each scoreset gets a dedicated view with the full score distribution as a histogram, the active calibration's LOF / intermediate / functional thresholds shaded in, and a calibration dropdown (IGVF Coding Variant Focus Group controls vs. study-provided cutoffs) that re-bands the histogram in place. URN, variant count, calibration count, and target gene shown in the header; one click opens the scoreset on MaveDB.

2026.05.06

Minor release

May 6, 2026

AlphaGenome predictions

Run Google DeepMind's AlphaGenome from the variant and gene pages and view the results inline. Predictions are computed live — first runs take 1–10 minutes, subsequent requests are cached.

AlphaGenome scores for rs7412 — High Impact classification with a tissue-by-gene expression-change heatmap across APOC2, APOC4, MARK4, and other nearby genes. — Variant page → Regulatory → AlphaGenome for rs7412.

Platform

NewVariant. Regulatory → AlphaGenome on any variant page scores the variant across seven scorers (regulatory disruption, DNA folding, expression change, gene activation, RNA splicing, RNA processing, splice site) and renders reference vs. alternate signal for CAGE, DNase, ATAC, RNA-seq, ChIP histone, ChIP TF, splicing, and contact maps. Filter by tissue group before predicting tracks.
NewGene. Cell/Tissue → AlphaGenome on any gene page returns region expression scores. AlphaGenome operates on fixed windows (≥512kb here), so any other gene inside the window also appears as a row in the heatmap — there is no single-gene scorer yet, read the row for your gene.

2026.04.29

Minor release

April 29, 2026

Disease and drug pages

Every disease and drug now has its own page. Search takes you straight there.

Acute Myeloid Leukemia disease page — clinical group, ClinGen and GenCC verdicts, prevalence, key phenotypes, and synonyms. — Disease — Acute Myeloid Leukemia.

Metformin drug page — clinical phase, approval year, black box warning, mechanism, and inline 2D structure. — Drug — Metformin.

Platform

NewDisease. Cross-references across MONDO, EFO, OMIM, Orphanet, MeSH, and DOID; clinical group; rare-disease flag; ClinGen and GenCC verdicts; prevalence and incidence; key phenotypes; full list of synonyms. Sections for genes (split into confirmed, implicated, and associated), drugs that treat it, GWAS variants, studies, phenotypes, and parent and sibling diseases — each labelled with how many it has so you know what's there before you open it.
NewDrug. Clinical phase, approval year, FDA flags such as black box warnings, mechanism of action, and the molecule rendered inline. Sections for targets, indications, pharmacogenomics, adverse effects, and drug-drug interactions — each labelled with how many it has so you know what's there before you open it.

Search

FixedCmd-K. Results are restricted to things that have a real page (genes, variants, diseases, drugs, phenotypes, cCREs). No more clicking a hit and landing somewhere dead.

2026.04.28

Minor release

April 28, 2026

Search history

The search palette now remembers what you've looked at. Recent searches appear as soon as you focus the input; pin the ones you keep coming back to.

Search

NewCmd-K. Recent searches show in the empty-state dropdown, synced across your devices when signed in. ↑↓ to navigate, Enter to open, ⌘-click to open in a new tab, ⌫ to remove. Hover a row to pin it to the top.

2026.04.09

Patch

April 9, 2026

GPN-MSA PHRED fix

gpn_msa_phred in v2026.02.15 was wrong. Recomputed across the full genome. Raw gpn_msa_score is unchanged.

Data

FixedGPN-MSA. gpn_msa_phred is now a CADD-style rank PHRED. Variants are sorted by gpn_msa_score ascending (lower = more deleterious), then phred = -10 * log10(rank / N), where rank counts variants at or below each score. Tied scores get the same PHRED.

2026.02.15

Major release

February 15, 2026

Missense, splicing, and non-coding pathogenicity expansion

A major expansion of variant effect prediction across protein-coding, splicing, and non-coding genome. Adds the full dbNSFP v5.2 missense suite, splicing and experimental assays, four non-coding pathogenicity scores, and gnomAD v4 regional constraint.

Data

NewdbNSFP v5.2. Full missense prediction suite. REVEL, BayesDel, ClinPred, MPC, ESM-1b, MetaRNN, MetaLR, M-CAP, VEST4, PrimateAI, DEOGEN2, MutPred2, MVP, gMVP, SIFT4G, PROVEAN, PhACTboost, MutFormer, VARITY, LIST-S2, PolyPhen-2 HumDiv/HumVar, MutationTaster, and MutationAssessor. 20+ predictors exposed under the dbnsfp struct.
NewAlphaMissense. Calibrated pathogenicity scores for every possible missense substitution, with per-transcript predictions and class labels (likely benign / ambiguous / likely pathogenic).
NewSpliceAI. Donor and acceptor gain/loss delta scores and positions, plus a per-variant max delta score for splicing impact.
NewMaveDB. Experimental variant effect scores from multiplexed assays of variant effect (MAVEs), linked by HGVS protein notation and scoreset URN.
NewGPN-MSA. Genome-wide pathogenicity from the Genomic Pre-trained Network trained on multiple sequence alignments. Raw and PHRED-scaled.
NewJARVIS. Non-coding variant importance and significance reclassifier. Raw and PHRED-scaled.
NewReMM. Regulatory Mendelian Mutation score for non-coding variants. Raw and PHRED-scaled.
NewNCBoost. Non-coding variant pathogenicity with linked-gene assignment and region classification.
NewMACIE. Multi-task classifier for non-coding pathogenicity with conserved, regulatory, and combined any-class predictions.
NewCV2F. Tissue-specific functional scores from MPRA and liver variant assays — liver, baseline, MPRA, LVS, and combined classifiers.
NewgnomAD constraint (Gnocchi). gnomAD v4 regional mutational constraint scores. Raw and PHRED-scaled, surfacing depleted-variation regions across the non-coding genome.
NewNCER. Non-Coding Essential Region percentile, ranking sites by predicted essentiality.
NewpgBoost. Probabilistic SNP-gene link scores from a gradient boosting model trained on multiome fine-mapping data, with percentile and linked gene.

2026.06.02

Minor release

June 2, 2026

Knowledge graph refresh

Platform

NewGene. New Functional Partners view under Gene Annotation puts gene-to-gene relations in one sortable, filterable table: co-expression (CoXPresdb z-score), co-essentiality, genetic interactions, TF regulation, and protein-protein interactions.
NewGene. Expression view adds single-cell expression by cell type (CellxGene Census) next to bulk tissue expression. Sort by specificity, mean expression, or fraction of cells expressing, and group by organ system.
NewGene. Protein Structure view now resolves the gene's protein product: UniProt identity, isoform transcripts (Ensembl), and protein-complex memberships (EBI Complex Portal), each with an outbound link.
NewGene. Disease Portfolio adds a Pharmacogenomic Guidelines table and a variant-centric prioritized targets table (genes prioritized for a disease or phenotype).
RemovedDetail pages. Disease pages drop the Studies tab. Phenotype pages drop the Variants and Studies tabs. The graph relations behind them were retired.
FixedGene. Pathway-leverage map is populated again. Parent and child pathways and member genes were coming back empty after the schema change.
FixedGraph Explorer. Edge sort and per-edge filters work again. The schema response shape changed and the controls were reading stale field names.

Data

New New node types: Protein, Transcript, Complex, Motif, Dataset, Scoreset, and CellLine. They appear in graph traversals now; dedicated pages come later.
New Gene-to-gene relations: co-expression (CoXPresdb v8), co-essentiality, genetic interactions, and TF regulation of target genes (TRRUST v2).
New Single-cell gene expression by cell type, from CellxGene Census 2024.
New Allele-specific disruption of TF binding by variants, from AdAstra.
New Drug-to-drug structural similarity, from ChEMBL fingerprints (Tanimoto).
New Variant-centric gene prioritization for diseases and phenotypes, plus pharmacogenomic guidelines.
New Gene links to its protein, transcripts, complexes, perturbation datasets, and DMS scoresets.
Updated Variant-trait associations were consolidated into a single relation (they used to be split into disease, phenotype, and EFO-trait buckets). Gene-disease scoring now keys on the Open Targets composite score (ot_score), with validity tier, ClinGen and GenCC classifications, and per-source evidence subscores.
Removed Retired the GWAS-Study and EFO-trait (Entity) node types, along with the study-investigates-trait, variant-reported-in-study, and per-bucket gene and variant trait relations.