AlphaGenome Atlas Deep-Linking & URL Configuration
Construct and validate deep-links for the AlphaGenome Atlas web application
(https://deepmind.google.com/science/alphagenome/atlas).
Base URL: https://deepmind.google.com/science/alphagenome/atlas
[!IMPORTANT] Mandatory Atlas Deep-Linking with Variant Scores: Whenever
presenting, discussing, or scoring genetic variants, you MUST always provide
clickable deep-links to the
AlphaGenome Atlas.
Use scripts/alphagenome_atlas_links.py to automate link and table
generation.
Prerequisites
# 1. Single Variant Exploration Link:
uv run scripts/alphagenome_atlas_links.py variant "chr9:128225994:G>A" \
--biosample K562 \
--modalities RNA_SEQ,DNASE,CHIP_TF
# 2. Genomic Locus / Interval Link:
uv run scripts/alphagenome_atlas_links.py locus "chr11:5288500-5290500" \
--biosample K562 \
--modalities RNA_SEQ,DNASE,CHIP_TF
# 3. Format Candidate Variant Records Table (with embedded clickable links):
uv run scripts/alphagenome_atlas_links.py table --input top_variants.json --biosample K562
# 4. Construct Ref vs. Alt Track Predictions Link (/atlas/track-predictions):
uv run scripts/alphagenome_atlas_links.py track-predictions \
--variant "chr15:42387805:C>G" \
--gene CAPN3 \
--biosample "Muscle_Skeletal"
2. URL Query Parameters
q (string, Required): Primary search target. Supports 1-based
closed intervals (chr11:5288500-5290500), gene symbols (BRCA1), Ensembl
IDs (ENSG00000012048), or 1-based variants (chr7:27170000:A>G).
m (enum, Optional): View mode. Defaults to entity for
genes/variants and locus for coordinate intervals. Use variant for
variant queries. (Allowed: locus, entity, variant, motifs).
i (string, Optional): Centered viewport zoom interval in 1-based
closed chr:start-end format (e.g. chr11:5289310-5289690). Required for
automatic motif rendering.
f (string, Optional): Comma-separated filter predicates in
KEY:VALUE format (e.g.
BIOSAMPLE_NAME:K562,SCORER_MODALITY:RNA-seq,ASSAY_TRANSCRIPTOR_FACTOR:GATA1).
Controls visible heatmap rows.
lItems (string, Optional): Layout item sequence, AVI score track
toggle (avi), section heatmaps, and pinned tracks list (e.g.
avi,section:RNA_SEQ,section:DNASE,pinned:<TrackKey>).
scores (string, Optional): Comma-separated list of ScoreId tokens
for the /atlas/track-predictions page comparison (e.g.
<ScoreId1>,<ScoreId2>).
md (enum, Optional): Active modality tab selector on the track
predictions view (RNA_SEQ, SPLICE_JUNCTIONS, SPLICE_SITE_USAGE,
DNASE).
tpRenames (string, Optional): Custom title overrides for specific
score predictions (ScoreId:CustomTitle).
tpLegendTitle (string, Optional): Custom legend title for the track
predictions chart card (e.g. Predicted Gene Expression).
[!IMPORTANT] Variant Query Format: Variants in q must strictly use
chr:pos_1_based:ref>alt format (e.g. chr7:27170000:A>G or URL-encoded
chr7:27170000:A%3EG, where the position is 1-based). Do not use
colon-separated alleles (A:G) or dbSNP rsIDs (rsIDs are unsupported).
3. Multi-Modality Filtering & The Assay Group Gotcha (f)
Filter Groups & Boolean Evaluation
Filters in f map to three primary evaluation groups:
Biosample Group (BIOSAMPLE_NAME, BIOSAMPLE_TYPE): Evaluated with
AND logic.
Assay Group (SCORER_MODALITY, ASSAY_TRANSCRIPTOR_FACTOR,
ASSAY_HISTONE_MARK): Evaluated with OR logic.
Gene Group (GENE_NAME): Evaluated with OR logic.
⚠️ Mandatory Multi-Modality Filter Rule
RNA-seq and DNase tracks have no transcription factor code
(transcriptionFactorCode === ""). If f contains only
ASSAY_TRANSCRIPTOR_FACTOR filters under the Assay group, RNA-seq and DNase
tracks fail the Assay evaluation and are hidden from the heatmap.
To display RNA-seq and DNase tracks alongside specific ChIP-seq
transcription factors, explicitly include SCORER_MODALITY:RNA-seq and
SCORER_MODALITY:DNase in f (handled automatically by
scripts/alphagenome_atlas_links.py):
f=BIOSAMPLE_NAME:<CellLine>,SCORER_MODALITY:RNA-seq,SCORER_MODALITY:DNase,ASSAY_TRANSCRIPTOR_FACTOR:<TF1>,ASSAY_TRANSCRIPTOR_FACTOR:<TF2>
4. Layout Configuration, AVI Scores, & Pinned Tracks (lItems)
Plotting AVI Scores and Modality Sections
- AVI Variant Impact Track (
avi): Including avi in lItems renders
the top-level AlphaGenome Variant Impact score track for the interval or
variant.
- Database Modality Sections (
section:<MODALITY>): Sections render full
unpinned heatmaps across all matching tracks for that modality (e.g.
section:RNA_SEQ, section:DNASE, section:CHIP_TF, section:ATAC,
section:CAGE).
Pinned Tracks & Motif Instances
[!NOTE] Track-Specific Motif Guideline: Pinned Active-ISM tracks with
motif instances and Contribution Weight Matrix (CWM) logos should only be
added when specifically requested for individual tracks. Only a limited
subset of tracks (such as key ChIP-TF or RNA-seq tracks relevant to the locus)
support and benefit from pinned motif overlays. For standard exploration
links, default section heatmaps
(avi,section:RNA_SEQ,section:DNASE,section:CHIP_TF) without pinned tracks
are preferred.
Motif instances and CWM logos render exclusively on pinned sequence tracks
(ACTIVE_ISM or RAW_SCORES, not REF_SCORES) at base-pair resolution.
General section heatmaps do not trigger motif footprint rendering.
Pinned Track Key Schema
pinned:<TrackMetadataName>:<StrandNumber>:<ScorerShortName>:<Source>
Example lItems entry in a raw URL query string:
pinned:EFO:0002067%20TF%20ChIP-seq%20GATA1:3:CHIP_TF:ACTIVE_ISM
<TrackMetadataName>: Exact track name from metadata proto (e.g.
EFO:0002067 TF ChIP-seq GATA1), percent-encoded (%20 for spaces) in raw
URL query strings (or unencoded when passed through
urllib.parse.urlencode).
<StrandNumber>: 1 (STRAND_POSITIVE), 2 (STRAND_NEGATIVE), 3
(STRAND_UNSTRANDED).
<ScorerShortName>: RNA_SEQ, CHIP_TF, DNASE, ATAC, CAGE,
PROCAP, CHIP_HISTONE (or RAW_SCORES-only modalities: SPLICE_SITES,
SPLICE_SITE_USAGE, SPLICE_JUNCTIONS, POLYADENYLATION, CONTACT_MAPS).
<Source>: Heatmap tileset source identifier: ACTIVE_ISM
(activity-weighted _TIMES_ACTIVE motif footprints and CWM sequence logos),
RAW_SCORES (raw ISM variant-score tracks, also rendering raw ISM motif
footprints and CWM sequence logos across all modalities), or REF_SCORES
(continuous reference signal line charts without sequence letters or
motifs). Calibrated and AVI tracks cannot be pinned, and legacy
:heatmap:...:<TilesetId> suffixes fail track lookup and are ignored.
Recipe for Automatic Motif Display on Load
- Append entries of the form
pinned:<TrackMetadataName>:<StrandNumber>:<ScorerShortName>:<Source>
(using ACTIVE_ISM for activity-weighted motifs or RAW_SCORES for raw ISM
motifs) directly to lItems for the specific target tracks only.
- Set viewport interval
i to base-pair resolution ($\ge 2\text{ px/bp}$,
window $\le 380\text{ bp}$).
- Configure
f with cell line and transcription factors.
5. Track Predictions & Ref vs. Alt Comparisons (/atlas/track-predictions)
The dedicated /atlas/track-predictions page compares predicted functional
profiles between the Reference and Alternate alleles for selected scores across
genomic windows:
- Route:
https://deepmind.google.com/science/alphagenome/atlas/track-predictions
- Visualizations: Expanded line plots (expression, chromatin
accessibility, TF binding) and Sashimi arc charts (splice junctions).
Automated Prediction Link Generation (scripts/alphagenome_atlas_links.py track-predictions)
Always construct track prediction URLs using scripts/alphagenome_atlas_links.py track-predictions. Manual ScoreId string formatting is error-prone due to
donor/acceptor skipping coordinates, strand orientation (+/-), and genic vs.
non-genic suffix rules. The script automatically handles coordinate extraction
from GENCODE v46, track catalog resolution, and URL synthesis.
# Variant & Gene:
uv run scripts/alphagenome_atlas_links.py track-predictions \
--variant "chr15:42387805:C>G" \
--gene CAPN3 \
--biosample "Muscle_Skeletal" \
--modalities SPLICE_JUNCTIONS,RNA_SEQ,DNASE,CHIP_TF \
--tf CTCF
# Interval/Locus query:
uv run scripts/alphagenome_atlas_links.py track-predictions \
--variant "chr15:42387805:C>G" \
--interval "chr15:41869312-42917888" \
--biosample "Muscle_Skeletal" \
--modalities SPLICE_JUNCTIONS,RNA_SEQ,DNASE,CHIP_TF
Supported CLI Options for track-predictions
--variant, -v (string, default: None): Variant string in
chr:pos_1_based:ref>alt format.
--gene, -g (string, default: None): Target gene symbol (bounds
i= viewport and computes splice junctions).
--gene_id (string, default: None): Target Ensembl gene ID (e.g.
ENSG00000092529.26).
--interval, -i (string, default: None): Genomic interval
viewport in chr:start-end format.
--biosample, -b (string, default: Muscle_Skeletal): Target
biosample or tissue query (e.g. Muscle_Skeletal, K562, Whole_Blood).
--modalities, -m (string, default:
SPLICE_JUNCTIONS,RNA_SEQ,DNASE,CHIP_TF): Comma-separated list of
modalities (SPLICE_JUNCTIONS, RNA_SEQ, DNASE, ATAC, CHIP_TF).
--tf (string, default: CTCF): Transcription factor name for
ChIP-TF tracks (e.g. CTCF, GATA1).
--rename (string, default: None): Custom track rename overrides in
the chart card.
--legend_title (string, default: None): Custom legend header for
the chart card.
--format (enum, default: table): Output format (table, url,
json).
[!IMPORTANT] Mandatory Splicing & RNA-seq Co-Plotting Rule: When
generating /atlas/track-predictions deep-links, plotting, or visualizing
variant impact data for splicing variants, always plot continuous RNA-seq
expression alongside splicing tracks (SPLICE_JUNCTIONS,
SPLICE_SITE_USAGE, SPLICE_SITES). Splicing mutations frequently activate
cryptic splice junctions and trigger nonsense-mediated decay (NMD) or alter
total transcript output; assessing splice junctions (sashimi arcs) together
with continuous RNA-seq read coverage is required to observe both the
structural splice defect and the resulting change in overall transcript
abundance.
[!IMPORTANT] Always Provide Bounded i= in Track Prediction URLs:
Omitting scores= or leaving the genomic interval (i=) unbounded causes the
web application to attempt querying all matching tracks across the broader
locus, leading to severe latency or page hanging. alphagenome_atlas_links.py track-predictions automatically bounds i= to the target gene or requested
interval.
1---2name: alphagenome-atlas-website-links-23description: "Constructs deep-links and URLs for the AlphaGenome Atlas website. Supports generating single-variant exploration links (1-based VCF-style coordinates), genomic locus views (1-based closed chr:start-end), candidate summary tables, and AlphaGenome reference vs. alternate predictions. Use when visualizing, exploring, charting, or linking genetic variants and genomic loci on the AlphaGenome Atlas, or when asked to inspect, view, or link predictions for a genomic variant. Don't use for computing raw AlphaGenome variant predictions directly without Atlas URLs (use alphagenome-single-variant-analysis or alphagenome-variant-impact-score)."4metadata:5 source: https://skillmd.com/skills/fridrichmethod/alphagenome-atlas-website-links-26---78# AlphaGenome Atlas Deep-Linking & URL Configuration910Construct and validate deep-links for the AlphaGenome Atlas web application11(`https://deepmind.google.com/science/alphagenome/atlas`).1213Base URL: `https://deepmind.google.com/science/alphagenome/atlas`1415> [!IMPORTANT] **Mandatory Atlas Deep-Linking with Variant Scores**: Whenever16> presenting, discussing, or scoring genetic variants, you **MUST always provide17> clickable deep-links to the18> [AlphaGenome Atlas](https://deepmind.google.com/science/alphagenome/atlas)**.19> Use `scripts/alphagenome_atlas_links.py` to automate link and table20> generation.2122--------------------------------------------------------------------------------2324## Prerequisites2526```bash27# 1. Single Variant Exploration Link:28uv run scripts/alphagenome_atlas_links.py variant "chr9:128225994:G>A" \29 --biosample K562 \30 --modalities RNA_SEQ,DNASE,CHIP_TF3132# 2. Genomic Locus / Interval Link:33uv run scripts/alphagenome_atlas_links.py locus "chr11:5288500-5290500" \34 --biosample K562 \35 --modalities RNA_SEQ,DNASE,CHIP_TF3637# 3. Format Candidate Variant Records Table (with embedded clickable links):38uv run scripts/alphagenome_atlas_links.py table --input top_variants.json --biosample K5623940# 4. Construct Ref vs. Alt Track Predictions Link (/atlas/track-predictions):41uv run scripts/alphagenome_atlas_links.py track-predictions \42 --variant "chr15:42387805:C>G" \43 --gene CAPN3 \44 --biosample "Muscle_Skeletal"45```4647--------------------------------------------------------------------------------4849## 2. URL Query Parameters5051* **`q`** (*string*, **Required**): Primary search target. Supports 1-based52 closed intervals (`chr11:5288500-5290500`), gene symbols (`BRCA1`), Ensembl53 IDs (`ENSG00000012048`), or 1-based variants (`chr7:27170000:A>G`).54* **`m`** (*enum*, Optional): View mode. Defaults to `entity` for55 genes/variants and `locus` for coordinate intervals. Use `variant` for56 variant queries. (Allowed: `locus`, `entity`, `variant`, `motifs`).57* **`i`** (*string*, Optional): Centered viewport zoom interval in 1-based58 closed `chr:start-end` format (e.g. `chr11:5289310-5289690`). Required for59 automatic motif rendering.60* **`f`** (*string*, Optional): Comma-separated filter predicates in61 `KEY:VALUE` format (e.g.62 `BIOSAMPLE_NAME:K562,SCORER_MODALITY:RNA-seq,ASSAY_TRANSCRIPTOR_FACTOR:GATA1`).63 Controls visible heatmap rows.64* **`lItems`** (*string*, Optional): Layout item sequence, AVI score track65 toggle (`avi`), section heatmaps, and pinned tracks list (e.g.66 `avi,section:RNA_SEQ,section:DNASE,pinned:<TrackKey>`).67* **`scores`** (*string*, Optional): Comma-separated list of `ScoreId` tokens68 for the `/atlas/track-predictions` page comparison (e.g.69 `<ScoreId1>,<ScoreId2>`).70* **`md`** (*enum*, Optional): Active modality tab selector on the track71 predictions view (`RNA_SEQ`, `SPLICE_JUNCTIONS`, `SPLICE_SITE_USAGE`,72 `DNASE`).73* **`tpRenames`** (*string*, Optional): Custom title overrides for specific74 score predictions (`ScoreId:CustomTitle`).75* **`tpLegendTitle`** (*string*, Optional): Custom legend title for the track76 predictions chart card (e.g. `Predicted Gene Expression`).7778> [!IMPORTANT] **Variant Query Format**: Variants in `q` must strictly use79> `chr:pos_1_based:ref>alt` format (e.g. `chr7:27170000:A>G` or URL-encoded80> `chr7:27170000:A%3EG`, where the position is 1-based). **Do not use**81> colon-separated alleles (`A:G`) or dbSNP rsIDs (rsIDs are unsupported).8283--------------------------------------------------------------------------------8485## 3. Multi-Modality Filtering & The Assay Group Gotcha (`f`)8687### Filter Groups & Boolean Evaluation8889Filters in `f` map to three primary evaluation groups:9091* **`Biosample` Group (`BIOSAMPLE_NAME`, `BIOSAMPLE_TYPE`)**: Evaluated with92 **AND** logic.93* **`Assay` Group (`SCORER_MODALITY`, `ASSAY_TRANSCRIPTOR_FACTOR`,94 `ASSAY_HISTONE_MARK`)**: Evaluated with **OR** logic.95* **`Gene` Group (`GENE_NAME`)**: Evaluated with **OR** logic.9697### ⚠️ Mandatory Multi-Modality Filter Rule9899`RNA-seq` and `DNase` tracks have no transcription factor code100(`transcriptionFactorCode === ""`). If `f` contains *only*101`ASSAY_TRANSCRIPTOR_FACTOR` filters under the Assay group, `RNA-seq` and `DNase`102tracks fail the Assay evaluation and are hidden from the heatmap.103104To display `RNA-seq` and `DNase` tracks alongside specific ChIP-seq105transcription factors, **explicitly include `SCORER_MODALITY:RNA-seq` and106`SCORER_MODALITY:DNase`** in `f` (handled automatically by107`scripts/alphagenome_atlas_links.py`):108109```110f=BIOSAMPLE_NAME:<CellLine>,SCORER_MODALITY:RNA-seq,SCORER_MODALITY:DNase,ASSAY_TRANSCRIPTOR_FACTOR:<TF1>,ASSAY_TRANSCRIPTOR_FACTOR:<TF2>111```112113--------------------------------------------------------------------------------114115## 4. Layout Configuration, AVI Scores, & Pinned Tracks (`lItems`)116117### Plotting AVI Scores and Modality Sections118119* **AVI Variant Impact Track (`avi`)**: Including `avi` in `lItems` renders120 the top-level AlphaGenome Variant Impact score track for the interval or121 variant.122* **Database Modality Sections (`section:<MODALITY>`)**: Sections render full123 unpinned heatmaps across all matching tracks for that modality (e.g.124 `section:RNA_SEQ`, `section:DNASE`, `section:CHIP_TF`, `section:ATAC`,125 `section:CAGE`).126127### Pinned Tracks & Motif Instances128129> [!NOTE] **Track-Specific Motif Guideline**: Pinned Active-ISM tracks with130> motif instances and Contribution Weight Matrix (CWM) logos should **only be131> added when specifically requested for individual tracks**. Only a limited132> subset of tracks (such as key ChIP-TF or RNA-seq tracks relevant to the locus)133> support and benefit from pinned motif overlays. For standard exploration134> links, default section heatmaps135> (`avi,section:RNA_SEQ,section:DNASE,section:CHIP_TF`) without pinned tracks136> are preferred.137138Motif instances and CWM logos render **exclusively on pinned sequence tracks**139(`ACTIVE_ISM` or `RAW_SCORES`, not `REF_SCORES`) at base-pair resolution.140General section heatmaps do not trigger motif footprint rendering.141142### Pinned Track Key Schema143144```145pinned:<TrackMetadataName>:<StrandNumber>:<ScorerShortName>:<Source>146```147148Example `lItems` entry in a raw URL query string:149150```151pinned:EFO:0002067%20TF%20ChIP-seq%20GATA1:3:CHIP_TF:ACTIVE_ISM152```153154* `<TrackMetadataName>`: Exact track name from metadata proto (e.g.155 `EFO:0002067 TF ChIP-seq GATA1`), percent-encoded (`%20` for spaces) in raw156 URL query strings (or unencoded when passed through157 `urllib.parse.urlencode`).158* `<StrandNumber>`: `1` (`STRAND_POSITIVE`), `2` (`STRAND_NEGATIVE`), `3`159 (`STRAND_UNSTRANDED`).160* `<ScorerShortName>`: `RNA_SEQ`, `CHIP_TF`, `DNASE`, `ATAC`, `CAGE`,161 `PROCAP`, `CHIP_HISTONE` (or `RAW_SCORES`-only modalities: `SPLICE_SITES`,162 `SPLICE_SITE_USAGE`, `SPLICE_JUNCTIONS`, `POLYADENYLATION`, `CONTACT_MAPS`).163* `<Source>`: Heatmap tileset source identifier: `ACTIVE_ISM`164 (activity-weighted `_TIMES_ACTIVE` motif footprints and CWM sequence logos),165 `RAW_SCORES` (raw ISM variant-score tracks, also rendering raw ISM motif166 footprints and CWM sequence logos across all modalities), or `REF_SCORES`167 (continuous reference signal line charts without sequence letters or168 motifs). Calibrated and AVI tracks cannot be pinned, and legacy169 `:heatmap:...:<TilesetId>` suffixes fail track lookup and are ignored.170171### Recipe for Automatic Motif Display on Load1721731. Append entries of the form174 `pinned:<TrackMetadataName>:<StrandNumber>:<ScorerShortName>:<Source>`175 (using `ACTIVE_ISM` for activity-weighted motifs or `RAW_SCORES` for raw ISM176 motifs) directly to `lItems` for the specific target tracks only.1772. Set viewport interval `i` to base-pair resolution ($\ge 2\text{ px/bp}$,178 window $\le 380\text{ bp}$).1793. Configure `f` with cell line and transcription factors.180181--------------------------------------------------------------------------------182183## 5. Track Predictions & Ref vs. Alt Comparisons (`/atlas/track-predictions`)184185The dedicated `/atlas/track-predictions` page compares predicted functional186profiles between the Reference and Alternate alleles for selected scores across187genomic windows:188189* **Route**:190 `https://deepmind.google.com/science/alphagenome/atlas/track-predictions`191* **Visualizations**: Expanded line plots (expression, chromatin192 accessibility, TF binding) and Sashimi arc charts (splice junctions).193194### Automated Prediction Link Generation (`scripts/alphagenome_atlas_links.py track-predictions`)195196Always construct track prediction URLs using `scripts/alphagenome_atlas_links.py197track-predictions`. Manual `ScoreId` string formatting is error-prone due to198donor/acceptor skipping coordinates, strand orientation (+/-), and genic vs.199non-genic suffix rules. The script automatically handles coordinate extraction200from GENCODE v46, track catalog resolution, and URL synthesis.201202```bash203# Variant & Gene:204uv run scripts/alphagenome_atlas_links.py track-predictions \205 --variant "chr15:42387805:C>G" \206 --gene CAPN3 \207 --biosample "Muscle_Skeletal" \208 --modalities SPLICE_JUNCTIONS,RNA_SEQ,DNASE,CHIP_TF \209 --tf CTCF210211# Interval/Locus query:212uv run scripts/alphagenome_atlas_links.py track-predictions \213 --variant "chr15:42387805:C>G" \214 --interval "chr15:41869312-42917888" \215 --biosample "Muscle_Skeletal" \216 --modalities SPLICE_JUNCTIONS,RNA_SEQ,DNASE,CHIP_TF217```218219### Supported CLI Options for `track-predictions`220221* **`--variant`, `-v`** (*string*, default: `None`): Variant string in222 `chr:pos_1_based:ref>alt` format.223* **`--gene`, `-g`** (*string*, default: `None`): Target gene symbol (bounds224 `i=` viewport and computes splice junctions).225* **`--gene_id`** (*string*, default: `None`): Target Ensembl gene ID (e.g.226 `ENSG00000092529.26`).227* **`--interval`, `-i`** (*string*, default: `None`): Genomic interval228 viewport in `chr:start-end` format.229* **`--biosample`, `-b`** (*string*, default: `Muscle_Skeletal`): Target230 biosample or tissue query (e.g. `Muscle_Skeletal`, `K562`, `Whole_Blood`).231* **`--modalities`, `-m`** (*string*, default:232 `SPLICE_JUNCTIONS,RNA_SEQ,DNASE,CHIP_TF`): Comma-separated list of233 modalities (`SPLICE_JUNCTIONS`, `RNA_SEQ`, `DNASE`, `ATAC`, `CHIP_TF`).234* **`--tf`** (*string*, default: `CTCF`): Transcription factor name for235 ChIP-TF tracks (e.g. `CTCF`, `GATA1`).236* **`--rename`** (*string*, default: `None`): Custom track rename overrides in237 the chart card.238* **`--legend_title`** (*string*, default: `None`): Custom legend header for239 the chart card.240* **`--format`** (*enum*, default: `table`): Output format (`table`, `url`,241 `json`).242243> [!IMPORTANT] **Mandatory Splicing & RNA-seq Co-Plotting Rule**: When244> generating `/atlas/track-predictions` deep-links, plotting, or visualizing245> variant impact data for splicing variants, **always plot continuous RNA-seq246> expression alongside splicing tracks** (`SPLICE_JUNCTIONS`,247> `SPLICE_SITE_USAGE`, `SPLICE_SITES`). Splicing mutations frequently activate248> cryptic splice junctions and trigger nonsense-mediated decay (NMD) or alter249> total transcript output; assessing splice junctions (sashimi arcs) together250> with continuous RNA-seq read coverage is required to observe both the251> structural splice defect and the resulting change in overall transcript252> abundance.253254> [!IMPORTANT] **Always Provide Bounded `i=` in Track Prediction URLs**:255> Omitting `scores=` or leaving the genomic interval (`i=`) unbounded causes the256> web application to attempt querying all matching tracks across the broader257> locus, leading to severe latency or page hanging. `alphagenome_atlas_links.py258> track-predictions` automatically bounds `i=` to the target gene or requested259> interval.