Working with spectra¶

Spectra is the lazy, iterable collection of Spectrum objects hanging off every file:

with mscompress.read("data.msz") as f:
    f.spectra              # Spectra
    f.spectra[0]           # Spectrum (first)
    f.spectra[42]          # Spectrum (42nd)
    len(f.spectra)         # int
    for s in f.spectra:    # iteration is in scan order
        ...

Spectrum properties¶

Property	Type	Notes
`index`	int	Position in the file
`scan`	int	Original scan number
`ms_level`	int	1, 2, ...
`retention_time`	float	seconds
`size`	int	Number of m/z–intensity pairs
`mz`	`numpy.ndarray`	1-D, dtype depends on source encoding
`intensity`	`numpy.ndarray`	1-D
`peaks`	`numpy.ndarray`	(N, 2) stack of m/z and intensity
`raw_mz`	bytes	Compressed/encoded raw bytes from the file
`raw_intensity`	bytes	Same, for intensity
`xml`	str	Spectrum XML element

mz, intensity, and peaks are computed lazily on first access — opening a 10 GB MSZ file is cheap; iterating its spectra without touching mz is cheap; reading s.mz triggers exactly one block decompression.

Transforms¶

Attach a callable to apply per-spectrum filtering or scaling at read time:

def threshold(spectrum):
    mask = spectrum.intensity > 1e3
    return spectrum.mz[mask], spectrum.intensity[mask]

f.spectra.set_transform(threshold)
for s in f.spectra:
    mz, intensity = s.peaks  # already filtered

with_transform() returns a new view without mutating the parent:

filtered = f.spectra.with_transform(threshold)
unfiltered = f.spectra

Iteration patterns¶

# All spectra
for s in f.spectra:
    ...

# Just MS2
ms2 = (s for s in f.spectra if s.ms_level == 2)

# A scan range
for i in range(1000, 2000):
    s = f.spectra[i]

For complex selections, prefer the extraction guide — it pushes the filter down into the C core.