mzpy 0.7.2

A pandas-like toolkit for MS data handling

mzpy: a pandas-like toolkit for MS data handling

=======

mzpy is a lightweight Python toolkit for mass spectrometry data processing with a focus on MS/MS (fragmentation) analysis. It provides:

• A NumPy-based container for spectra (MSdata)
• Utilities for m/z matching and de-duplication
• Robust MS/MS similarity scoring (Bonanza, cosine variants, entropy)
• A Pandas-like PeakFrame for handling MSP/MGF-style records, matching spectra in batch, and exporting/importing formats
• Precursor type helpers for adduct/loss inference

The library borrows and re-implements ideas from MS-DIAL scoring for consistent, reproducible analysis.

Features at a glance

• MSdata: ndarray subclass to hold spectra as [mz, intensity], with normalization, centroiding, basic filtering.
• m/z utils: absolute and ppm matching, array matching, and deduplication.
• Similarity: Numba-accelerated alignment and multiple similarity metrics (matched peaks, Bonanza, cosine, modified cosine, spectral entropy, reverse dot).
• Batch scoring: compute multiple similarity matrices efficiently.
• PeakFrame: a Pandas DataFrame subclass to parse, normalize, match, filter, export, and visualize MS/MS peak lists.
• Precursor type: enumerate plausible adduct/loss types and m/z for a given exact mass.

Installation

• From PyPI:

  • pip install mzpy

• From source (editable):

  • git clone https://github.com/<your-user>/mzpy.git
  • cd mzpy
  • pip install -e .

Requirements (key):

• numpy, pandas
• numba
• rdkit (for some precursor type utilities)
• Optional: matplotlib (for plotting chromatograms via PeakFrame.plot_chrom)

Quick start

1) Represent spectra with MSdata

import numpy as np
from mzpy.ms import MSdata

# Create from an array of [mz, intensity]
arr = np.array([
    [100.0, 10.0],
    [101.0, 50.0],
    [150.5, 5.0],
])
ms = MSdata(arr, to_normalized=True)  # normalizes intensity to 0–100
print(ms.max_intensity_mz)  # m/z at max intensity
print(ms.mz)                # m/z column

MSdata includes:

• normalize(inpalce=True|False)
• filter_out(threshold=...)
• centroid(...) for peak picking
• insert_precursormz(mz) to prepend a precursor
• max_intensity_mz, max_mz, mz, intensity properties

2) m/z matching utilities

import numpy as np
from mzpy import mz

# Single value matching
mz.match(100.0000, 100.0025, tol=0.003, is_abs_error=True)  # True if within 0.003 Da

# Arrays matching
que = np.array([100.0, 101.0, 150.5])
ref = np.array([100.002, 101.1, 150.5])
que_unmatched, que_matched, ref_unmatched, ref_matched = mz.match_array(que, ref, tol=0.003)

# Count matches
n = mz.match_num(que, ref, tol=0.003)

# Deduplicate m/z values
uniq = mz.unique([100.0, 100.002, 101.0], tol=0.003)

3) Compute similarity between two spectra

The similarity module aligns two MS/MS spectra then computes multiple metrics.

import numpy as np
from mzpy import similarity

# Join precursor and fragments to form a spectrum (first row is precursor)
ms_left  = similarity.join(precursormz=300.1234, msms=[[100.1, 50], [150.2, 100], [200.3, 30]])
ms_right = similarity.join(precursormz=300.1235, msms=[[100.1, 45], [150.2, 95],  [199.9, 20]])

# Alignment tolerance: tol=(precursor_tol_da, fragment_tol_da)
scores = similarity.get_scores_weighted(
    ms_left, ms_right, tol=(0.003, 0.005), method="weighted_average"
)
print(scores)
# {
#   'matched_peaks_ratio': ...,
#   'spectral_entropy': ...,
#   'modified_dot_product': ...,
#   'bonanza_score': ...,
#   'reverse_dot_product': ...,
#   'simple_dot_product': ...,
#   'total_score': ...
# }

Under the hood:

• align(...) (Numba) aligns two spectra by precursor and fragment tolerance.
• Score functions are Numba-accelerated for speed:
  • get_matched_num, get_matched_peaks_score
  • get_bonanza_score
  • get_simple_dot_product (cosine)
  • get_modified_dot_product_score
  • get_entropy_similarity
  • get_reverse_dot_product

For batch comparisons (matrix form):

# Prepare a list of spectra, each from `similarity.join`
query_list = [
    similarity.join(300.1234, [[100.1, 50], [150.2, 100]]),
    similarity.join(305.1111, [[120.1, 70], [160.2, 80]]),
]
ref_list = [
    similarity.join(300.1236, [[100.1, 55], [150.2, 90]]),
]

matched_mx, ratio_mx, bonanza_mx, simple_mx, modified_mx, entropy_mx = \
    similarity.get_scores_batch(query_list, ref_list, tol=(0.003, 0.005))

4) PeakFrame: Work with MSP/MGF-like datasets

PeakFrame extends pandas.DataFrame for MS/MS records with an MSMS column as a list of [mz, intensity].

Load an MSP file exported by MS-DIAL (v5.2+) or any MSP in the supported structure:

from mzpy.peak import PeakFrame, read_msp, read_mgf

# MSP from MS-DIAL (ensure the format matches PeakFrame._parse_msp_txt rules)
df = read_msp("test/example.msp", sep_ms2="\t")  # test folder has example MSP files

# Or read MGF
mgf = read_mgf("test/example.mgf", sep_ms2=" ")

# Update Num Peaks automatically
df.update_num_peaks()
print(df.head())

Match and score spectra between two PeakFrame objects (or within the same object):

# Self-match to get pairwise similarities
scores_long = df.match(tol=(0.003, 0.005))
print(scores_long.head())
# columns include:
#   idx, que_idx, matched_counts, matched_ratio, bonanza, simple_dot, modified_dot, entropy

# Match against another PeakFrame
df2 = read_msp("test/example_ref.msp")
scores_long2 = df.match(que=df2, tol=(0.003, 0.005))

Drop likely duplicate MS/MS entries using similarity thresholds:

dedup = df.drop_duplicated_ms(
    mz_on="precursormz",
    MSMS_on="MSMS",
    tol=(0.003, 0.005),
    sim_thd={'bonanza': 0.9, 'entropy': 0.9, 'matched_ratio': 0.25},
    keep_first_on=None
)

Export MSP:

df.to_msp("out.msp", standardized=True, MSMS_sep="\t", chunk_size=5000)

Other utilities:

• flatten_msms_mz(...): collect all fragment m/z into a flat array
• extract_ion_chrom(...): simple EIC extraction based on MS1 precursor m/z window
• find_precursor_type(target_mass, ionmode, mz_on): infer plausible precursor types and annotate rows
• IO helpers: read_pickle, read_sql, to_pickle, to_sqlite3

5) Precursor type suggestions

from mzpy.precursorType import load_precursors

# Enumerate precursor candidates for exact mass in a specific ion mode
candidates = load_precursors(exact_mw=300.1234, mode='pos')  # or 'neg'
print(candidates[['type', 'ionmode', 'charge', 'mz']].head())

Use with PeakFrame:

annot = df.find_precursor_type(target_mass=300.1234, ionmode='pos', mz_on='precursormz')

Data format notes

• MS/MS (MSMS) is a list (or ndarray) of pairs: [mz, intensity].
• Many functions normalize intensity to relative abundance (0–100).
• Tolerances:
  • Precursor: typically absolute Da tolerance (e.g., 0.003 Da).
  • Fragments: typically absolute Da tolerance (e.g., 0.005 Da).
  • ppm-based matching for single m/z values is supported via mz.match(..., is_abs_error=False).

Examples and test data

• The test folder contains example .msp files for quick experiments and demonstrations of the PeakFrame readers.
• Try:
  • df = read_msp("test/your_example.msp")
  • df.match(tol=(0.003, 0.005))

API reference (selected)

• mzpy.ms
  • MSdata(array, metadata=None, to_normalized=True)
  • .centroid(...), .normalize(...), .filter_out(...), .insert_precursormz(...)
  • Properties: .mz, .intensity, .max_intensity_mz, .max_mz
• mzpy.mz
  • match(mz1, mz2, tol=0.003, is_abs_error=True)
  • match_array(que, ref, tol=0.003)
  • match_num(que, ref, tol=0.003)
  • unique(mz_values, tol=0.003)
• mzpy.similarity
  • join(precursormz, msms, intensity=100)
  • join_array(mz_list, msms_list, intensity=100)
  • align(ms_left, ms_right, tol_precursor, tol_fragment) [Numba]
  • Score functions (all Numba): get_bonanza_score, get_simple_dot_product, get_modified_dot_product_score, get_entropy_similarity, get_reverse_dot_product, get_matched_num, get_matched_peaks_score
  • get_scores(ms_left, ms_right, tol0, tol1) [Numba; returns array]
  • get_scores_weighted(..., method="weighted_average") [Python aggregation]
  • get_scores_batch(que_list, ref_list=None, tol=(...))
• mzpy.peak (PeakFrame)
  • Readers: read_msp, read_mgf, read_msd_msp, read_mona_msp, read_pickle, read_sql
  • Matching: match, drop_duplicated_ms, match_counts, match_by_chunk_to_csv (experimental)
  • Utilities: update_num_peaks, round_msms, flatten_msms_mz, extract_ion_chrom, find_precursor_type
  • Export: to_msp, to_pickle, to_sqlite3
• mzpy.precursorType
  • load_precursors(exact_mw, mode='pos'|'neg')

Performance notes

• Core alignment and scoring functions are Numba-jitted for speed.
• For very large tables, consider chunked matching (match_by_chunk_to_csv) or pre-filter by precursor m/z proximity to reduce pair counts.

Limitations and tips

• Ensure MS/MS arrays do not contain NaN; normalization routines will raise errors.
• Reverse dot product implementation notes in code comment indicate it may need review for specific analytical goals.
• MSP parsing assumes simple “Key: Value” headers and two-column peak lines. For non-standard MSPs, you may adapt parsing or use read_mona_msp.

License

GNU General Public License v3.0

Citation

If you use mzpy in academic work, please cite the repository:

• GitHub: https://github.com/zq-lab/mzpy