Discovery & Browsing

Powering proteome-wide exploration with a confidence-weighted global impact engine.

To enable researchers to navigate the vast landscape of isoform structural perturbations, SPLISOFORMS utilizes a dedicated Discovery Index. This engine pre-calculates a multi-dimensional Structural Impact Score (SIS) for every isoform, facilitating real-time discovery of high-disruption events across the entire proteome.

The Structural Impact Score (SIS)

The SIS is a confidence-weighted additive model: each biological event type contributes a point score, and the sum is multiplied by the mean pLDDT of the structural model. This ensures that structurally unreliable predictions contribute proportionally less to the final score.

Step 1 — Raw Score
raw = Δlen + Ddomain + Nneo + Bnmd + Pptm
Step 2 — Confidence Weighting
SIS = raw × (pLDDT / 100)

pLDDT acts as a confidence multiplier (0–1). Low-quality structural models contribute proportionally less to the final score, preventing false high-impact calls from poorly modelled sequences.

Length Delta
min(|ΔL| / 20, 5)
max 5 pts

Large sequence deviations indicate major structural overhauls. Capped at 5 to prevent outlier dominance.

Domain Impact
2 × lost + 0.5 × gained
unbounded

Lost Pfam domains are weighted twice as much as gained ones — loss of conserved function is the most reliable indicator of biological disruption.

Neoantigens
min(count × 2, 4)
max 4 pts

Each predicted MHC-binding peptide adds 2 points. Capped at 4 to avoid score inflation from over-predicted isoforms.

NMD Status
3 (target) / 1 (suspect) / 0
max 3 pts

Transcripts predicted to be degraded by Nonsense-Mediated Decay receive a strong penalty, as they likely never produce protein.

PTM Alterations
1.5 (lost) + 0.5 (exposed) + 0.5 (buried)
max 2.5 pts

Regulatory site deletions (lost) are weighted higher than accessibility changes (exposed/buried) that still retain the site.

Scale: Typical scores range from 0 to ~15. An isoform with a large sequence delta (+100 aa), two lost domains, predicted neoantigens, and NMD targeting would score ~10–14 before confidence weighting. Scores above 10 generally indicate high-priority candidates for experimental follow-up.

Impact Explorer Interface

The Results page exposes the Discovery Index through two complementary browse interfaces, toggled from the page header:

Isoform Browser

Proteome-wide isoform ranking by Structural Impact Score. Filter by impact profile, pLDDT, and isoform density; click any row to open the full isoform detail page.

Epitope Atlas

Proteome-wide neoantigen table with deduplication, sortable columns, and filters for HLA allele, PTM tier, binding category, junction type, and pLDDT. See the Neoantigens section for details.

Key components of the Isoform Browser:

Summary Cards
Four top-level cards show aggregate counts: domain-loss isoforms, PTM-disrupted isoforms, neoantigenic isoforms, and NMD-targeted transcripts across the entire knowledgebase.
Quick-Stats Strip
A four-tile strip below the summary cards shows database-wide averages: mean |ΔSeq| (aa), mean domains lost per isoform, total NMD targets, and total PTM site losses.
Impact Profile Filters
All:No filter applied — shows all isoforms sorted by impact score.
High Confidence:Only isoforms with pLDDT ≥ 70; structural predictions are reliable.
NMD Targets:Transcripts flagged as NMD target or suspect — likely not translated.
Domain Loss:Isoforms where at least one Pfam domain is lost relative to canonical.
Neoantigens:Isoforms with at least one predicted MHC-binding peptide.
PTM Disrupted:Isoforms that delete or alter the accessibility of a PTM site.
Sliders & Sorting
Filter by minimum pLDDT confidence and minimum isoforms per gene. Sort by impact score, pLDDT, or delta length in either direction. All filters apply instantly via the pre-indexed table.

Discovery Index Architecture

Because on-the-fly calculation across 52,000+ isoforms is computationally prohibitive, we maintain a pre-indexed flattened table (isoform_discovery_index) optimized for sub-second retrieval. The index is populated by a backfill script and must be re-run whenever the underlying data or scoring formula changes.

High Performance
Aggregate subqueries and specialized indexing allow the Impact Explorer to filter 50k+ records in under 100ms with real-time sorting and pagination.
Multi-modal
Unifies data from structural models, Pfam HMMs, NMD predictions, PTM accessibility, and MHC binding into a single analytical row per isoform.
← Prev: Drug Interactions
Back to StartOverview