MSProfile.bin

Binary payload file containing profile-mode (full-scan) mass spectra. Only present when the instrument was configured to record profile data. In all corpus files tested, if MSProfile.bin exists with non-zero size, the corresponding MSScan.bin scan records have a SpectrumFormatID=1 block pointing into it.

Status: Confirmed

Two block encodings occur. Every profile block written by acquisition software B.05 to B.09 in the corpus is an LZF block; every block written by version 10.1 is a run-length block (see Evidence). The reader tells them apart per block and converts the axis with the scan's mass calibration (see 05-msmasscal.md).

File Layout

• Global header: 68 bytes. The first 4 bytes are an int32 magic number/version flag: 258 (0x02 0x01 0x00 0x00). The remaining 64 bytes are all zeros padding.
• Data payload: Contiguous scan blocks, each located by SpectrumOffset / ByteCount in the scan's SpectrumFormatID=1 block in MSScan.bin. UncompressedByteCount is 16 + PointCount * 4 in both encodings.

Decoded Layout

Field	Type
x0: flight time of point 0	f64
dx: sampling period	f64
PointCount intensities	integer counts

• Point i is at flight time x0 + i * dx: a uniform time axis, not a uniform m/z axis.
• Flight times are converted to m/z exactly like Q-TOF centroid x-values (03-mspeak.md), with the scan's calibration from 05-msmasscal.md.

LZF Blocks

• The block is a raw LZF stream (no frame header) starting at byte 0. There is no uncompressed preamble.
• It decompresses to exactly UncompressedByteCount bytes, little-endian:

Bytes	Field	Type
0-7	x0	f64
8-15	dx	f64
16-	PointCount intensities	int32

Run-Length Blocks

Little-endian throughout:

Bytes	Field	Type
0-7	x0	f64
8-15	dx	f64
16-19	marker 00 ff 02 90	bytes
20-23	minus the index of the first encoded point	int32
24-	stream	signed integers of varying width

The stream is read as signed integers of the current width, which starts at 4 bytes. Starting at the point given by bytes 20-23, for each integer v:

• v >= 0: v is the intensity of the current point; move to the next point.
• v < 0: let u = -v - 1 (the bitwise complement of v). Skip u >> 2 points and set the width to 1 << (u & 3) bytes (u & 3 = 0, 1, 2, 3 for 1, 2, 4, 8 bytes).

Points the stream does not reach are zero, including those before the first encoded point and after the last. The stream ends with the block; the last point it reaches is at most PointCount.

A run of zeros and a change of width share one negative integer. The first integer of the stream sets the width of the first values; in the corpus it is -1 (1-byte values) in 189,405 blocks and -2 (2-byte values) in 3,098.

The reader detects this encoding by the marker at bytes 16-19 following a valid axis header (x0 > 0, dx > 0, both finite); any other block is decoded as LZF.

Evidence

LZF. The first 16 bytes of a PXD001310 profile block are:

07 00 00 00  00 80 d6 d4  40 20 06 20  00 01 e0 3f

Read as LZF, these bytes are a literal run of 8 bytes (0x07), two back-references (0x20 0x06, 0x20 0x00) and a 2-byte literal (0x01). They decompress to 16 bytes: x0 = 21338.0 and dx = 0.5. A unit test in crates/openaraw/src/raw/msprofile.rs decodes them.

Corpus check (qtof_calibration_check example): every profile block in the 26 LZF profile bundles (PXD001310, PXD027071, PXD038602, PXD038606, PXD049393, PXD057471; acquisition software B.05 to B.09) decodes. Where the bundle's summary fields describe profile data, the sum of the int32 intensities equals the stored TIC, the most intense point equals BasePeakValue, and that point's calibrated m/z equals BasePeakMZ (worst-bundle p95 0.00045 ppm).

Run-length. The first block of PXD044639/A1FU.d starts:

00 00 00 00 00 d2 d4 40  00 00 00 00 00 00 e0 3f   x0 = 21320.0, dx = 0.5
00 ff 02 90                                        marker
18 be ff ff                                        -16872: first point 16872
ff ff ff ff                                        -1: skip 0, width 1
04 df 06 ef 01 16 19 fe 83 00 ff ff 4e ...

From point 16872: 04 is 4; df (-33, u = 32) skips 8 points; 06 is 6; ef (-17) skips 4; 01 16 19 are 1, 22, 25; fe (-2, u = 1) switches to 2-byte values without a skip; 83 00 is 131; ff ff (-1) switches back to 1-byte values; 4e is 78. A unit test in crates/openaraw/src/raw/msprofile.rs decodes this prefix.

Corpus check (tests/profile_run_length.rs): all 192,503 blocks in the 13 PXD044639 bundles (G6550B, AcqSoftwareVersion "6200 series TOF/6500 series Q-TOF 10.1 (48.0)") decode. For every scan the decoded intensities sum exactly to the stored TIC, their maximum equals BasePeakValue exactly, and a point holding that maximum, converted with the DefaultMassCal.xml calibration, is within 0.0002 ppm of BasePeakMZ. 1-, 2- and 4-byte values occur; 8-byte values do not. Bytes 16-19 are 00 ff 02 90 in every block.

Unresolved

• The meaning of the run-length marker bytes (00 ff 02 90); they are constant across the corpus and are only used to detect the encoding.
• 8-byte values (u & 3 = 3) follow from the 2-bit width code but are not observed in the corpus.