PTM Accessibility
Quantifying how splicing rewires PTM accessibility — with confidence masking and a relative-change threshold so calls reflect genuine conformational change, not noise.
Post-translational modifications (PTMs) are chemical marks that switch a protein's activity, stability, or localisation on and off. A modification can only occur if its target residue is physically reachable — so when splicing reshapes the fold around a PTM site, it can quietly enable or disable that regulatory switch. For every known PTM site on the canonical isoform, SPLISOFORMS projects the residue onto the novel isoform through a splice-aware alignment (the same k-mer-anchored algorithm used by the MSA panel) and recomputes its solvent-accessible surface area (SASA) with FreeSASA on both AF3 models.
Sites come from each gene's reviewed UniProt entry, and we track the three best-annotated regulatory modifications: phosphorylation (on Ser/Thr/Tyr), ubiquitination, and acetylation (both on Lys). UniProt records many rarer modification types, but these three dominate the annotated regulatory landscape and have the coverage needed for reliable comparison across isoforms.
Five-way classification
A site is reclassified only when both the absolute Δ (> 10 Ų) and the relative Δ (> 25% of the canonical SASA) clear their thresholds, and when the minimum pLDDT across the two structures is ≥ 70. Otherwise the call stays at "intact" or "undetermined" — we'd rather be honest about uncertainty than inflate the count of regulatory rewirings.
| Class | Condition | What it means |
|---|---|---|
| Intact | |Δ SASA| < 10 Ų OR |Δ rel| < 25% | Site retained with no meaningful accessibility change. |
| Exposed | Δ SASA > 10 Ų AND Δ rel > +25% | Site moved into a more solvent-accessible state — both the absolute and relative thresholds must clear. |
| Buried | Δ SASA < -10 Ų AND Δ rel < -25% | Site moved into a more solvent-shielded state. A conformational reshuffle or domain insertion. |
| Lost | Canonical residue deleted | Splice-aware alignment maps the canonical position to a gap in the novel — the residue is not in the protein at all. |
| Undetermined | Missing SASA OR min pLDDT < 70 | SASA could not be computed for one or both structures, OR the residue lies in a low-confidence region of the AF prediction. Side-chain placement is unreliable here — we deliberately refuse to call exposed/buried. |
Why a relative threshold matters
Baseline SASA varies an order of magnitude across residues — a fully-buried residue can have ~5 Ų, a fully-exposed surface residue ~150 Ų. An absolute ±10 Ų rule labels the same physical shift as "huge" in one case and "noise" in the other. Requiring a simultaneous ±25% relative change calibrates the call to the residue's own baseline. The combined rule was chosen because:
- Either threshold alone over-calls at the extremes — a relative-only rule inflates changes on tiny buried residues, an absolute-only rule inflates them on large exposed ones. Requiring both keeps only shifts that are real in absolute and relative terms.
- Undetermined is kept as its own class, separate from intact, so that "we couldn't measure this reliably" is never silently reported as "no change".
Reading the lollipop map
On each isoform page, PTM sites are drawn as a lollipop map along the protein sequence (aa 1 on the left to the C-terminus on the right), in two stacked tracks:
- Canonical reference — every known site, coloured by modification type (phosphorylation, ubiquitination, acetylation).
- Isoform accessibility — the same sites at their canonical positions, coloured by their accessibility call (exposed, buried, or unchanged); sites spliced out of the isoform appear as an ✕. A coverage ribbon above the track marks which stretches of the canonical protein are present, deleted, or outside the isoform.
Newly exposed
A phosphosite buried in the canonical fold ends up on the surface of the isoform — a kinase that couldn't reach it now can.
Newly buried
A site exposed in the canonical is shielded in the isoform — the modification is effectively blocked.
Lost (✕)
The modified residue is spliced out of the isoform entirely — the regulatory site is gone.
Structural context (TED)
We also tag each site by its position relative to the TED structural-domain boundaries. This is contextual only — it does not change the accessibility classification, and it is drawn purely from the structure-based TED domains, not from Pfam sequence domains.
Intra-domain
The site sits within a stable, folded domain in the novel isoform.
Disrupted
The site was inside a canonical domain that splicing fractured — no longer in a domain in the novel.
Inter-domain
The site is in a flexible linker / loop region in both isoforms, OR no TED domains were annotated.