Rationale and technical details of ProteoformTracker's design¶

In TDP or MDP, the question "Given two (or more) specific proteoforms, can this mass spectrometer actually tell them apart?" might sound straightforward. However, it is more complex than relying on a single delta mass cutoff. Three things a naive fixed-threshold check misses, and that this tool handles explicitly instead:

1. The mass difference needed for resolvability isn't fixed¶

It scales with the pair's own mass and charge state, via the instrument's resolving power at whatever m/z the ions actually land at:

\[ R(m/z) = R_{ref} \times \sqrt{\frac{m/z_{ref}}{m/z}} \qquad\qquad \Delta M_{FWHM} \approx \frac{M}{R(m/z_{best})} \]

A rule like "Δmass ≥ 3 Da is enough" produces false positives on large pairs (a Δmass that clears the fixed threshold can still be small relative to the actual peak width at high m/z, where resolving power is worse — the rule calls them separable when they aren't) and false negatives on small, marginal pairs (a Δmass under the threshold can already be cleanly resolved at low m/z, where resolving power is far better — the rule calls them unresolvable when they're actually fine). See Resolving power & FWHM for the full model and a worked example.

2. A resolvable pair can still be confounded by a third, unrelated protein¶

Even if your two proteoforms of interest are cleanly separable from each other, some completely unrelated protein — from a different gene entirely — might share one of their intact masses, or land a charge-state peak on top of one of theirs. ProteoformTracker searches the full reference proteome for this explicitly (see Relevant vs. confounding proteins), rather than treating the pair as isolated from the rest of the proteome.

This is also the reasoning behind treating the MS2 fragment ladder as a first-class, independent axis of discrimination rather than an optional extra: two proteins from different genes essentially never share extended sequence identity, so their fragment ions diverge quickly even when their intact masses happen to coincide. See Fragmentation propensity & scoring modes.

3. Getting there requires real structural and physical modeling, not just arithmetic¶

None of the above is a lookup or a plain mass subtraction — each part rests on its own nontrivial model:

For a proteoform derived from a novel transcript (an rMATS event, or a pasted/uploaded FASTA sequence that isn't yet a named, annotated isoform), even knowing where it structurally differs from the canonical or other known isoforms has to be inferred — differential exons are found by matching against annotated transcript structure (see Option 2: FASTA sequence and Option 3: rMATS splicing events), not assumed from the input alone. Only once that's resolved can ProteoformTracker say which MS2 fragment ions are actually proteoform-unique rather than shared with the canonical sequence (see Tier colors).
Predicting whether two proteoforms would actually show up as separate MS1 peaks means modeling the charge-state envelope each one would realistically produce under the chosen ionization mode, and the isotope distribution within each charge state — resolved into discrete peaks or blurred into one smooth envelope, depending on the instrument (see Charge-state envelopes).
Distinguishing two proteoforms at MS2 means scoring each backbone bond's real fragmentation propensity — not just enumerating every theoretical fragment mass, since different bonds fragment at genuinely different rates depending on sequence context (see Fragmentation propensity & scoring modes).

How the pieces fit together¶

flowchart LR
    A[Input: gene/isoform, FASTA sequence, or rMATS event] --> B["Proteoform\n(sequence + PTMs)"]
    B --> C[MS1: charge-envelope\n+ resolving-power scoring]
    B --> D[MS2: fragment ladder\n+ propensity scoring]
    C --> E[Section 1:\nrelevant-proteoform comparison]
    D --> E
    B --> F[Confounder search\nmass + m/z domain]
    F --> G[Section 2:\nconfounding-protein search]

Every input path (gene/isoform selection, pasted FASTA, or an rMATS splicing event) resolves into the same internal Proteoform representation — a mature amino-acid sequence plus a list of applied PTMs — before any scoring runs. Everything downstream is identical regardless of which path produced it.

Known limitations¶

Stated here plainly, not hidden in a footnote — ProteoformTracker's own design spec is explicit about these:

The envelope-crowding check is a closed-form heuristic, not a guarantee of real deconvolution-algorithm (e.g. UniDec) behavior.
The fragmentation propensity model predicts cleavage likelihood, not observed ion intensity — no mature, generalizable intensity-prediction model exists yet for intact-protein fragmentation.
PTMs are propagated onto the proteoforms you explicitly specify, never onto background confounding proteins — PTM occupancy is sample/condition-specific biology, not a fixed proteome property, so there's no canonical "modified state" of an unrelated protein to precompute.
Only terminal (b/y) fragment ions are modeled; internal fragments (two simultaneous cleavages) are out of scope — they're combinatorially far more numerous and individually much less probable than terminal ions.
The Orbitrap R(m/z) scaling law is instrument-family specific; FT-ICR or other platforms would need a different resolving-power relationship.