Struct FeaturesConfig 

pub struct FeaturesConfig {

    pub min_charge: u8,
    pub max_charge: u8,
    pub min_chain_cosine: f64,
    pub min_isotope_step_ratio: f64,
    pub max_isotopes: usize,
    pub max_isotope_log2_ratio: f64,
    pub chain_predicted_intensity_gate: bool,
    pub min_scan_overlap: usize,
    pub exhaustive_min_isotope_score: f64,
    pub isotope_evidence_ratio_sigma: f64,
    pub isotope_evidence_cosine_shape: f64,
    pub exhaustive_isotope_priority: bool,
    pub cosine_anchor: String,
    pub sulfur_offsets: Vec<i8>,
    pub isotope_model: IsotopeModelSpec,
    pub neutron_mass: f64,
    pub min_isotope_score: f64,
    pub min_cosine_score: f64,
    pub min_combined_score: f64,
}

Fields

min_charge: u8

max_charge: u8

min_chain_cosine: f64

Per-extension chromatographic-cosine threshold used while building an isotope chain. The candidate hill’s cosine vs the seed hill must clear this to be added; otherwise chain extension stops in that direction.

min_isotope_step_ratio: f64

Minimum intensity a candidate isotope hill must retain relative to its chain predecessor, as a fraction: a heavier isotope must be at least this fraction of its predecessor’s intensity or the chain stops. Chain extension runs upward only (the seed IS the monoisotope hypothesis), so there is no downward counterpart to this knob. Default 0.01. The pre-0.3.0 name right_max_decrease is still accepted.

max_isotopes: usize

max_isotope_log2_ratio: f64

Per-extension intensity-ratio gate. After a candidate hill clears min_chain_cosine, also check that its apex intensity vs the predecessor’s matches the averagine ratio within ± this many log2 units. Catches column-bleed / contaminant hills that co-elute (so pass cosine) but have wrong intensity for an isotope. Set to a very large value to effectively disable. Typical: 1.5 (within ~2.83× of expected).

chain_predicted_intensity_gate: bool

When true (legacy behaviour), isotope-chain extension stops as soon as the averagine-predicted intensity of the next isotope falls below the per-run noise floor (5th-pct hill intensity × 0.8), computed from the seed intensity before the hill is searched for. This cheaply trims decayed chain tails, but it also blocks legitimate low-abundance monoisotopic seeds from ever pairing their M+1: a seed already near the floor has a predicted M+1 below the floor, so the chain never forms and the feature collapses to charge 0. When false (the default), this predicted-intensity break is skipped and chain extension is terminated purely by evidence: a missing hill (find_neighbors empty), min_chain_cosine, the max_isotope_log2_ratio intensity-ratio gate, min_isotope_step_ratio (found hill must be ≥ this fraction of the predecessor), and the averagine-template-length / max_isotopes caps. Setting this false recovers dim 2+/3+ features that have both isotope hills present but were never paired — recommended when feeding an MS1 search that needs feature depth. Re-benchmark FDR/quant when flipping.

min_scan_overlap: usize

Minimum number of mutually-overlapping scans two hills must share before a chromatographic cosine is computed between them; below it the cosine is 0 and the isotope chain will not extend across that pair. Default 3 (the long-standing hard gate — keep for PXD003881).

On fast gradients hills are only 3–5 scans wide, so a dim isotope hill frequently overlaps its monoisotope by just 1–2 scans and is rejected outright regardless of how well the shapes agree — the dominant cause of koth emitting a low-abundance monoisotope as an unpaired charge-0 feature on such data (biosaur2 requires only ≥1 shared scan). Lowering this to 2 recovers those pairs; pair it with a shorter hills.min_scans (a 2-scan hill can never reach a 3-scan overlap) and re-benchmark quant, since short-overlap pairs are noisier.

exhaustive_min_isotope_score: f64

Minimum isotope-pattern (Bhattacharyya) score a candidate — original or truncated — must reach before it may claim its hills; below-bar candidates are dropped so their hills stay free for a better-fitting feature. Default 0.0 = no gate.

isotope_evidence_ratio_sigma: f64

Signal standard deviation of seed-relative log2 isotope-ratio errors. The broad noise null has fixed sigma 2; valid range is 0 < sigma < 2.

isotope_evidence_cosine_shape: f64

Shape of the Beta(shape, 1) co-elution model against a uniform null. Must be finite and > 1; larger values favor tighter co-elution.

exhaustive_isotope_priority: bool

Legacy compatibility field; no longer affects claim ordering. The exhaustive resolver always ranks by additive isotope log evidence. Accepted so existing configuration files continue to deserialize.

cosine_anchor: String

Chromatographic-cosine anchor for isotope-chain extension: which hill each candidate isotope’s cosine gate is measured against.

"seed" (the default): anchor every isotope’s cosine to the monoisotope seed hill (the convention biosaur2 / AlphaPept / Dinosaur all use). The m/z step target and the intensity-ratio predecessor still step from the immediate predecessor — only the cosine reference changes. Rejects a far isotope that co-elutes with its neighbour but not with the mono. Beat "adjacent" on the 20-run PXD003881 cohort (+0.31 pp recall); see default_cosine_anchor.

"adjacent" (legacy): anchor to the immediate predecessor in the chain — for M+1 that is the seed, for M+k≥2 the previously-claimed isotope hill. koth’s former default; byte-identical to the pre-2026-07 paper output.

The per-extension cosine that feeds the composite mean_cosine is computed against whichever reference this selects (so the composite is consistent with the gate). The reported cosine_score field and the exhaustive-resolver rescoring remain adjacent-style regardless.

sulfur_offsets: Vec<i8>

Sulfur-count offsets to score each isotope chain against, relative to the ceiling of the averagine-expected sulfur count. The chain is scored against one averagine template per offset and the best Bhattacharyya kept — this corrects the systematic penalty on Cys/Met-rich peptides whose M+2 is elevated by ³⁴S (4.25 %, +2 Da).

Default [-1, 0, 1], i.e. {0, 1, 2} sulfurs up to 2665 Da, {1, 2, 3} to 5330 Da, {2, 3, 4} to 7995 Da. Negative results saturate at 0 and duplicates collapse, so [-2, -1, 0, 1] on a small peptide is {0, 1, 2}. Widen it (e.g. [-2, -1, 0, 1]) to keep the no-sulfur template on large peptides — ~32 % of 3 kDa tryptic peptides have no sulfur at all.

An empty list disables sulfur awareness: one plain averagine template at the half-up rounded count.

Scoring takes a max over templates, so a longer list can only raise scores — including for decoys and mis-assembled chains. Judge a change on discrimination (PSM recall / target–decoy separation), never on the isotope-score distribution alone.

isotope_model: IsotopeModelSpec

Which analyte class’s average composition the theoretical isotope pattern is built from.

"peptide" (the default) is Senko’s averagine and the only model the published benchmark exercises. "rna" and "dna" are the means of the four ribo- / deoxyribonucleotide chain residues. An explicit table overrides both:

isotope_model = { residue_mass = 321.2916, c = 9.5, h = 11.75, n = 3.75, o = 7.0 }

s defaults to 0 and a model without sulfur ignores sulfur_offsets entirely — varying a structurally absent atom would hand every candidate, decoys included, a free maximum over templates. An unknown class name is a config parse error, never a silent fallback to peptide.

neutron_mass: f64

Neutron (C13) mass in Da

min_isotope_score: f64

Drop features whose isotope_score (Bhattacharyya vs averagine) is below this. 0.0 = keep all. Default 0.5.

min_cosine_score: f64

Drop features whose cosine_score (mean chromatographic cosine of adjacent isotope hills) is below this. 0.0 = keep all.

min_combined_score: f64

Drop features whose combined_score (= isotope × cosine) is below this. 0.0 = keep all. All three filters are AND-ed.

Implementations

impl FeaturesConfig

pub fn validate(&self) -> Result<(), Error>

Validate distribution parameters and strings that serde alone cannot check. Called after TOML deserialization so invalid values fail at load time rather than silently falling back at runtime.

pub fn cosine_anchor_mode(&self) -> CosineAnchor

Resolve the chromatographic-cosine anchor. Infallible at runtime because validate() has already rejected bad values at config load; the defensive fallback preserves legacy behaviour if reached.

Trait Implementations

impl Clone for FeaturesConfig

fn clone(&self) -> FeaturesConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for FeaturesConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Default for FeaturesConfig

fn default() -> FeaturesConfig

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for FeaturesConfig

where FeaturesConfig: Default,

fn deserialize<__D>( __deserializer: __D, ) -> Result<FeaturesConfig, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for FeaturesConfig

fn serialize<__S>( &self, __serializer: __S, ) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.