chrisagrams/massive_kb_v1_shuffled

MassIVE-KB v1 — globally-shuffled parquet

Tandem mass spectra from the MassIVE-KB v1 peptide spectral library, repackaged as globally-shuffled, zstd-compressed Parquet shards for streaming model training.

Why this repackaging

The source MassIVE-KB MGF is ordered by source raw file, with the same peptide repeated in consecutive entries. Feeding that order to training leaves long runs of correlated spectra inside any locally-shuffled read window, which produces loss spikes. Here every spectrum is assigned to a uniformly-random shard in a single streaming pass, so each shard is an i.i.d. sample of the whole library and a row-group read from any shard is already a near-perfect global shuffle.

Splits

Split	Directory	Shards	Spectra	Size
train	train/	256	28,508,636	~47 GB
validation	val/	256	1,000,234	~1.9 GB
test	test/	256	996,027	~1.9 GB
total		768	30,504,897	~51 GB

Each split is independently sharded; shards within a split are interchangeable i.i.d. samples (no ordering meaning to the shard index).

Schema

Each row is one spectrum:

Column	Type	Description
m/z	list<float32>	Peak m/z values
int	list<float32>	Peak intensities (parallel to m/z)
peptide_charge	string	SEQUENCE_z — peptide sequence + precursor charge

Peptide notation

peptide_charge is the bare peptide sequence followed by _<charge>, e.g. EALEVAVAGTELFR_2. Modifications are inline, in bracket form with an explicit sign, attached to the residue (or at the N-terminus):

• M[+15.995] — oxidized methionine
• C[+57.021] — carbamidomethyl cysteine
• [-17.027]QSPLGFEVHDEITVTR_2 — N-terminal pyro-Glu (loss of ammonia)

Stripping the bracketed mods (re.sub(r"\[[+-][0-9.]+\]", "", pep)) recovers the bare amino-acid sequence. The precursor m/z is not stored; it is recomputed from SEQUENCE_z downstream.

Spectra without a sequence, charge, or peaks were dropped during conversion.

Usage

from datasets import load_dataset

# Stream a split without downloading all 51 GB
ds = load_dataset("chrisagrams/massive_kb_v1_shuffled", split="train", streaming=True)
ex = next(iter(ds))
print(ex["peptide_charge"], len(ex["m/z"]))

Or read the Parquet directly with pyarrow:

import pyarrow.parquet as pq
t = pq.read_table("train/train_000.parquet")

Provenance & license

Derived from the MassIVE-KB v1 peptide spectral library (https://massive.ucsd.edu/). Use of the underlying data is subject to MassIVE / GNPS terms; please cite MassIVE-KB and verify the applicable license for your use before redistribution. This repository only re-encodes and shuffles the spectra; no spectra were added or modified beyond the mod-notation rewrite described above.