thyra 4.6.0__tar.gz → 4.6.2__tar.gz

{thyra-4.6.0 → thyra-4.6.2}/PKG-INFO

@@ -1,11 +1,12 @@
 Metadata-Version: 2.5
 Name: thyra
-Version: 4.6.0
+Version: 4.6.2
 Summary: A modern Python library for converting Mass Spectrometry Imaging (MSI) data into SpatialData/Zarr format - your portal to spatial omics
 Project-URL: Homepage, https://github.com/M4i-Imaging-Mass-Spectrometry/thyra
 Project-URL: Repository, https://github.com/M4i-Imaging-Mass-Spectrometry/thyra
 Project-URL: Documentation, https://M4i-Imaging-Mass-Spectrometry.github.io/thyra/
-Project-URL: Paper, https://doi.org/10.1021/jasms.6c00169
+Project-URL: Paper, https://doi.org/10.1038/s41592-026-03305-5
+Project-URL: Paper (JASMS), https://doi.org/10.1021/jasms.6c00169
 Project-URL: Bug Tracker, https://github.com/M4i-Imaging-Mass-Spectrometry/thyra/issues
 Project-URL: Discussions, https://github.com/M4i-Imaging-Mass-Spectrometry/thyra/discussions
 Project-URL: Changelog, https://github.com/M4i-Imaging-Mass-Spectrometry/thyra/blob/main/CHANGELOG.md
@@ -223,7 +224,15 @@ MIT -- see [LICENSE](LICENSE).
 
 ## Citation
 
-If you use Thyra, please cite the paper:
+If you use Thyra, please cite:
+
+> Visvikis, T.; Hendriks, T. F. E.; Vierdag, W.-M.; Marconato, L.; Heeren, R. M. A.;
+> Cuypers, E. Thyra: bridging mass spectrometry imaging and SpatialData for
+> unified multimodal analysis. *Nat. Methods* **2026**.
+> [doi:10.1038/s41592-026-03305-5](https://doi.org/10.1038/s41592-026-03305-5)
+
+If your work relies on how Thyra reads vendor files or keeps memory bounded
+while converting, please also cite:
 
 > Visvikis, T.; Vierdag, W.-M.; Marconato, L.; Birmpili, A.; Heeren, R. M. A.;
 > Cuypers, E. Native, Memory-Bounded Conversion of Mass Spectrometry Imaging
@@ -231,39 +240,19 @@ If you use Thyra, please cite the paper:
 > *37* (10), 2379-2387.
 > [doi:10.1021/jasms.6c00169](https://doi.org/10.1021/jasms.6c00169)
 
-```bibtex
-@article{visvikis2026thyra,
-  title   = {Native, Memory-Bounded Conversion of Mass Spectrometry Imaging Data into the {SpatialData} Ecosystem},
-  author  = {Visvikis, Theodoros and Vierdag, Wouter-Michiel and Marconato, Luca and Birmpili, Angeliki and Heeren, Ron M. A. and Cuypers, Eva},
-  journal = {Journal of the American Society for Mass Spectrometry},
-  year    = {2026},
-  volume  = {37},
-  number  = {10},
-  pages   = {2379--2387},
-  doi     = {10.1021/jasms.6c00169}
-}
-```
-
-The same entry is in [CITATION.cff](CITATION.cff), which GitHub's "Cite this
-repository" button reads. To cite the software itself:
-
-```bibtex
-@software{thyra2024,
-  title = {Thyra: Modern Mass Spectrometry Imaging Data Conversion},
-  author = {Visvikis, Theodoros},
-  year = {2024},
-  url = {https://github.com/M4i-Imaging-Mass-Spectrometry/thyra}
-}
-```
+BibTeX for both is in [CITATION.bib](CITATION.bib). GitHub's "Cite this
+repository" button reads [CITATION.cff](CITATION.cff), which shows the Nature
+Methods paper and lists the JASMS paper under its references.
 
 ### Papers
 
 | Paper | Published | DOI |
 |---|---|---|
+| Thyra: bridging mass spectrometry imaging and SpatialData for unified multimodal analysis (*Nat. Methods*) | 2026-10 | [10.1038/s41592-026-03305-5](https://doi.org/10.1038/s41592-026-03305-5) |
 | Native, Memory-Bounded Conversion of Mass Spectrometry Imaging Data into the SpatialData Ecosystem (*J. Am. Soc. Mass Spectrom.*) | 2026-09 | [10.1021/jasms.6c00169](https://doi.org/10.1021/jasms.6c00169) |
 | Thyra: Bridging Mass Spectrometry Imaging and SpatialData for Unified Multi-Modal Analysis (bioRxiv preprint) | 2026-01 | [10.64898/2026.01.26.700792](https://doi.org/10.64898/2026.01.26.700792) |
 
-Ion mobility and MS/MS support came after both papers. If you use them,
+Ion mobility and MS/MS support came after the JASMS paper and the preprint. If you use them,
 please also name the Thyra version you used.
 
 ## Acknowledgments

{thyra-4.6.0 → thyra-4.6.2}/README.md

@@ -169,7 +169,15 @@ MIT -- see [LICENSE](LICENSE).
 
 ## Citation
 
-If you use Thyra, please cite the paper:
+If you use Thyra, please cite:
+
+> Visvikis, T.; Hendriks, T. F. E.; Vierdag, W.-M.; Marconato, L.; Heeren, R. M. A.;
+> Cuypers, E. Thyra: bridging mass spectrometry imaging and SpatialData for
+> unified multimodal analysis. *Nat. Methods* **2026**.
+> [doi:10.1038/s41592-026-03305-5](https://doi.org/10.1038/s41592-026-03305-5)
+
+If your work relies on how Thyra reads vendor files or keeps memory bounded
+while converting, please also cite:
 
 > Visvikis, T.; Vierdag, W.-M.; Marconato, L.; Birmpili, A.; Heeren, R. M. A.;
 > Cuypers, E. Native, Memory-Bounded Conversion of Mass Spectrometry Imaging
@@ -177,39 +185,19 @@ If you use Thyra, please cite the paper:
 > *37* (10), 2379-2387.
 > [doi:10.1021/jasms.6c00169](https://doi.org/10.1021/jasms.6c00169)
 
-```bibtex
-@article{visvikis2026thyra,
-  title   = {Native, Memory-Bounded Conversion of Mass Spectrometry Imaging Data into the {SpatialData} Ecosystem},
-  author  = {Visvikis, Theodoros and Vierdag, Wouter-Michiel and Marconato, Luca and Birmpili, Angeliki and Heeren, Ron M. A. and Cuypers, Eva},
-  journal = {Journal of the American Society for Mass Spectrometry},
-  year    = {2026},
-  volume  = {37},
-  number  = {10},
-  pages   = {2379--2387},
-  doi     = {10.1021/jasms.6c00169}
-}
-```
-
-The same entry is in [CITATION.cff](CITATION.cff), which GitHub's "Cite this
-repository" button reads. To cite the software itself:
-
-```bibtex
-@software{thyra2024,
-  title = {Thyra: Modern Mass Spectrometry Imaging Data Conversion},
-  author = {Visvikis, Theodoros},
-  year = {2024},
-  url = {https://github.com/M4i-Imaging-Mass-Spectrometry/thyra}
-}
-```
+BibTeX for both is in [CITATION.bib](CITATION.bib). GitHub's "Cite this
+repository" button reads [CITATION.cff](CITATION.cff), which shows the Nature
+Methods paper and lists the JASMS paper under its references.
 
 ### Papers
 
 | Paper | Published | DOI |
 |---|---|---|
+| Thyra: bridging mass spectrometry imaging and SpatialData for unified multimodal analysis (*Nat. Methods*) | 2026-10 | [10.1038/s41592-026-03305-5](https://doi.org/10.1038/s41592-026-03305-5) |
 | Native, Memory-Bounded Conversion of Mass Spectrometry Imaging Data into the SpatialData Ecosystem (*J. Am. Soc. Mass Spectrom.*) | 2026-09 | [10.1021/jasms.6c00169](https://doi.org/10.1021/jasms.6c00169) |
 | Thyra: Bridging Mass Spectrometry Imaging and SpatialData for Unified Multi-Modal Analysis (bioRxiv preprint) | 2026-01 | [10.64898/2026.01.26.700792](https://doi.org/10.64898/2026.01.26.700792) |
 
-Ion mobility and MS/MS support came after both papers. If you use them,
+Ion mobility and MS/MS support came after the JASMS paper and the preprint. If you use them,
 please also name the Thyra version you used.
 
 ## Acknowledgments

{thyra-4.6.0 → thyra-4.6.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "thyra"
-version = "4.6.0"
+version = "4.6.2"
 description = "A modern Python library for converting Mass Spectrometry Imaging (MSI) data into SpatialData/Zarr format - your portal to spatial omics"
 authors = [{ name = "Theodoros Visvikis", email = "t.visvikis@maastrichtuniversity.nl" }]
 maintainers = [{ name = "Theodoros Visvikis", email = "t.visvikis@maastrichtuniversity.nl" }]
@@ -241,14 +241,15 @@ dependencies = [
     # Floor is scikit-image's own, which is also comfortably above the 9.5
     # that `Image.has_transparency_data` needs; no upper bound, nothing here
     # has tested one.
-    "pillow>=10.1",                   # tested 12.3.0
+    "pillow>=10.1",                   # tested 13.0.0
 ]
 
 [project.urls]
 Homepage = "https://github.com/M4i-Imaging-Mass-Spectrometry/thyra"
 Repository = "https://github.com/M4i-Imaging-Mass-Spectrometry/thyra"
 Documentation = "https://M4i-Imaging-Mass-Spectrometry.github.io/thyra/"
-Paper = "https://doi.org/10.1021/jasms.6c00169"
+Paper = "https://doi.org/10.1038/s41592-026-03305-5"
+"Paper (JASMS)" = "https://doi.org/10.1021/jasms.6c00169"
 "Bug Tracker" = "https://github.com/M4i-Imaging-Mass-Spectrometry/thyra/issues"
 Discussions = "https://github.com/M4i-Imaging-Mass-Spectrometry/thyra/discussions"
 Changelog = "https://github.com/M4i-Imaging-Mass-Spectrometry/thyra/blob/main/CHANGELOG.md"

{thyra-4.6.0 → thyra-4.6.2}/thyra/__init__.py

@@ -70,7 +70,7 @@ warnings.filterwarnings(
     category=FutureWarning,
 )
 
-__version__ = "4.6.0"
+__version__ = "4.6.2"
 
 #: Public name -> the module that defines it, imported on first access.
 _LAZY_ATTRS = {

{thyra-4.6.0 → thyra-4.6.2}/thyra/converters/spatialdata/base_spatialdata_converter.py

@@ -43,6 +43,7 @@ log capture, which is the consumer that met it (issue #249).
 import asyncio
 import errno
 import logging
+import math
 import warnings
 from abc import ABC, abstractmethod
 from contextlib import contextmanager
@@ -637,6 +638,65 @@ class BaseSpatialDataConverter(BaseMSIConverter, ABC):
         """Apply :meth:`build_uns_metadata` to an AnnData about to be written."""
         self.uns.apply(adata, self._uns_context())
 
+    def _refuse_a_pitch_past_float(self) -> None:
+        """Refuse a pitch whose product with the grid is not a finite number.
+
+        A finite pitch passes every earlier check, but ``1e308`` um times
+        the third pixel is infinity: ``obs["spatial_x"]``, the pixel boxes
+        and the affines then held ``inf`` at exit 0, and ``thyra validate``
+        passed the store (issue #527). Checked once the grid and every
+        pitch are settled, so a detected pitch and a stated one, 2D and 3D,
+        all meet it. The far edge is the largest index plus half a pixel,
+        counted from the source's own origin, which the store also scales.
+        """
+        assert self._dimensions is not None
+        offsets = tuple(self._source_coordinate_offsets or (0, 0, 0)) + (0, 0, 0)
+        axes = [
+            ("x", "pixel size", self.pixel_size_um, 0),
+            ("y", "pixel size", self.pixel_size_y_um, 1),
+        ]
+        if self.handle_3d:
+            axes.append(("z", "z spacing", self.z_spacing_um, 2))
+        for axis, name, pitch, i in axes:
+            n = int(self._dimensions[i])
+            edge = (abs(float(offsets[i])) + n + 0.5) * float(pitch)
+            if not math.isfinite(edge):
+                raise ConversionRefused(
+                    f"A {name} of {pitch:g} um across {n:,} pixels along {axis} "
+                    "places the far edge of the image past the largest number "
+                    "a float holds. Give a pixel size in micrometres "
+                    "(--pixel-size, --z-spacing)."
+                )
+
+    def _refuse_a_grid_past_memory(self) -> None:
+        """Refuse a pixel grid whose per-position arrays will not fit.
+
+        Asked before the mass axis and the tables, so a grid of 10**12
+        positions for two spectra is refused in one line rather than found
+        by ``np.zeros`` (issue #527). Every plane is counted, as the
+        per-plane route may hold them all at once. The free memory read
+        here is the one the mass-axis check compares with too.
+
+        Raises:
+            ConversionRefused: When the projection exceeds the fraction of
+                free memory a conversion may take.
+        """
+        from ...resampling.memory_guard import available_memory_gb, pixel_grid_refusal
+
+        assert self._dimensions is not None
+        free = getattr(self, "_axis_free_memory_gb", None)
+        if free is None:
+            free = available_memory_gb()
+            self._axis_free_memory_gb = free
+        refusal = pixel_grid_refusal(
+            self._dimensions,
+            int(self._n_spectra or 0),
+            bool(getattr(self.reader, "has_acquisition_order", False)),
+            available_gb=free,
+        )
+        if refusal is not None:
+            raise ConversionRefused(refusal)
+
     def _initialize_conversion(self) -> None:
         """Override parent initialization to preserve resampled mass axis."""
         logger.info("Loading essential dataset information...")
@@ -670,6 +730,8 @@ class BaseSpatialDataConverter(BaseMSIConverter, ABC):
 
             # After pixel size, because the fallback is the pixel size.
             self._resolve_z_spacing(essential)
+            self._refuse_a_pitch_past_float()
+            self._refuse_a_grid_past_memory()
 
             # Handle mass axis setup
             self._setup_mass_axis()
@@ -1021,7 +1083,7 @@ class BaseSpatialDataConverter(BaseMSIConverter, ABC):
         )
         return unique.astype(indices.dtype, copy=False), summed
 
-    def build_region_numbers(self, x_values, y_values) -> NDArray[np.int32]:
+    def build_region_numbers(self, x_values, y_values) -> NDArray[np.int64]:
         """``obs["region_number"]`` for the given pixel positions, in row order.
 
         Written on every dataset for a consistent schema, so a consumer
@@ -1043,16 +1105,18 @@ class BaseSpatialDataConverter(BaseMSIConverter, ABC):
             y_values: Pixel y index per obs row.
 
         Returns:
-            Region number per obs row.
+            Region number per obs row, as int64: a source may number its
+            regions past 2**31, which an int32 column ended in an
+            ``OverflowError`` (issue #527).
         """
         n = len(x_values)
         region_map = getattr(self, "_region_map", None)
         if region_map is None:
-            return np.ones(n, dtype=np.int32)
+            return np.ones(n, dtype=np.int64)
 
         keys = zip(x_values.tolist(), y_values.tolist())
         return np.fromiter(
-            (region_map.get(key, -1) for key in keys), dtype=np.int32, count=n
+            (region_map.get(key, -1) for key in keys), dtype=np.int64, count=n
         )
 
     def _create_mass_dataframe(self) -> pd.DataFrame:

thyra-4.6.2/thyra/core/positions.py

@@ -0,0 +1,165 @@
+"""Position, index and region columns as whole numbers, or a refusal.
+
+Vendor files state pixel positions, spectrum numbers and region numbers in
+whatever type their writer chose: ``uint64`` in a Parquet column, ``REAL``
+or text in a sqlite one, a float a writer computed as ``0.3 / 0.1``. Each
+reader used to cast with ``astype(np.int64)`` or ``int()``, so a float just
+below a whole number was cut down onto its neighbour's pixel, a ``uint64``
+of ``2**63`` or more wrapped negative, and text or ``NULL`` ended in a
+traceback (issue #527).
+
+:func:`whole_numbers` is the one door they go through instead. A value is
+accepted when it is a whole number within int64 once floating-point noise
+is allowed for: a float within :data:`WHOLE_TOLERANCE` of a whole number is
+that number (``2.9999999999999996`` is 3), anything further off is not a
+position. Everything else is refused in one line that names the column,
+how many values fail and the first of them.
+
+The tolerance is for a column whose type is a float, as mzPeak allows for
+positions: there a computed value carries writer noise. A vendor sqlite
+column declared INTEGER that holds ``100.9999999`` is damage, not noise,
+and its readers pass ``tolerance=0``.
+"""
+
+import numbers
+from typing import Any, Optional
+
+import numpy as np
+from numpy.typing import NDArray
+
+from ..errors import ConversionRefused
+
+__all__ = ["WHOLE_TOLERANCE", "whole_number", "whole_numbers", "whole_sql"]
+
+#: How far a float may sit from a whole number and still be that number.
+#: Writer noise (``0.3 / 0.1``) is ~1e-15 of a pixel; a value 1e-6 away
+#: is still that pixel by any reading, and 0.5 away is no pixel at all.
+WHOLE_TOLERANCE = 1e-6
+
+_INT64_MAX = np.iinfo(np.int64).max
+_INT64_MIN = np.iinfo(np.int64).min
+
+
+def _refuse(what: str, source: Any, n_bad: int, n_all: int, first: Any) -> None:
+    where = f" in {source}" if source is not None else ""
+    raise ConversionRefused(
+        f"{n_bad} of {n_all} {what}{where} are not whole numbers within "
+        f"int64 (the first is {first!r}), so no pixel can be placed from them"
+    )
+
+
+def _from_objects(
+    values: NDArray[Any], what: str, source: Any, tolerance: float
+) -> NDArray[np.int64]:
+    """Python objects (a sqlite column): ints, floats, or something else."""
+    out = np.empty(values.size, dtype=np.int64)
+    bad = []
+    for i, value in enumerate(values.tolist()):
+        number: Optional[int] = None
+        if isinstance(value, bool):
+            number = None
+        elif isinstance(value, numbers.Integral):
+            number = int(value)
+        elif isinstance(value, numbers.Real):
+            number = _whole_float(float(value), tolerance)
+        if number is None or not _INT64_MIN <= number <= _INT64_MAX:
+            bad.append(value)
+            continue
+        out[i] = number
+    if bad:
+        _refuse(what, source, len(bad), values.size, bad[0])
+    return out
+
+
+def _whole_float(value: float, tolerance: float) -> Optional[int]:
+    if not np.isfinite(value):
+        return None
+    nearest = round(value)
+    if abs(value - nearest) > tolerance:
+        return None
+    return int(nearest)
+
+
+def whole_number(value: Any) -> Optional[int]:
+    """One value of a column declared INTEGER, or ``None`` when it is not whole.
+
+    For a reader that skips a bad record rather than refusing the file: an
+    ``int``, or a float with no fraction, within int64. Text, ``None``, a
+    bool and a float with any fraction at all give ``None``.
+    """
+    if isinstance(value, bool):
+        return None
+    if isinstance(value, numbers.Integral):
+        number = int(value)
+    elif isinstance(value, numbers.Real):
+        found = _whole_float(float(value), 0.0)
+        if found is None:
+            return None
+        number = found
+    else:
+        return None
+    return number if _INT64_MIN <= number <= _INT64_MAX else None
+
+
+def whole_sql(column: str) -> str:
+    """SQL that holds where a sqlite ``column`` stores a whole number.
+
+    An INTEGER, or a REAL with no fraction; text, ``NULL``, a REAL with a
+    fraction and a REAL past int64 fail it, the last because ``CAST``
+    saturates. The same rule as :func:`whole_number`, for a query that
+    should leave such rows out.
+    """
+    return f"({column} = CAST({column} AS INTEGER))"
+
+
+def whole_numbers(
+    values: Any, what: str, source: Any = None, tolerance: float = WHOLE_TOLERANCE
+) -> NDArray[np.int64]:
+    """``values`` as int64 whole numbers, or a refusal naming the first bad one.
+
+    Args:
+        values: A column of positions, indices or region numbers: any numpy
+            array or sequence, of any integer, float or object dtype.
+        what: What the values are, plural, as it should read mid-sentence --
+            ``"x positions"``, ``"spectrum indices"``.
+        source: The file or archive, named in the refusal.
+        tolerance: How far a float may sit from a whole number and still
+            be read as it; 0 for a column declared as an integer.
+
+    Returns:
+        The values as an int64 array of the same length.
+
+    Raises:
+        ConversionRefused: If a value is not a whole number within int64:
+            ``NULL``, text, ``nan``, ``inf``, a float further than
+            :data:`WHOLE_TOLERANCE` from a whole number, or an unsigned
+            value of ``2**63`` or more.
+    """
+    array = np.asarray(values)
+    if array.ndim != 1:
+        array = array.reshape(-1)
+    if array.size == 0:
+        return array.astype(np.int64)
+    kind = array.dtype.kind
+    if kind == "b":
+        _refuse(what, source, array.size, array.size, array.flat[0])
+    if kind == "i":
+        return array.astype(np.int64, copy=False)
+    if kind == "u":
+        over = np.nonzero(array > np.uint64(_INT64_MAX))[0]
+        if over.size:
+            _refuse(what, source, over.size, array.size, int(array[over[0]]))
+        return array.astype(np.int64)
+    if kind == "f":
+        as_float = array.astype(np.float64, copy=False)
+        nearest = np.rint(as_float)
+        with np.errstate(invalid="ignore"):
+            # 2**63 is exactly representable and already out of range;
+            # every float below it in magnitude fits.
+            bad = ~np.isfinite(as_float) | (np.abs(as_float - nearest) > tolerance)
+            bad |= (nearest >= 2.0**63) | (nearest < -(2.0**63))
+        if bad.any():
+            first = np.nonzero(bad)[0][0]
+            _refuse(what, source, int(bad.sum()), array.size, float(as_float[first]))
+        return nearest.astype(np.int64)
+    return _from_objects(array, what, source, tolerance)

{thyra-4.6.0 → thyra-4.6.2}/thyra/metadata/extractors/bruker_extractor.py

@@ -6,6 +6,7 @@ from pathlib import Path
 from typing import Any, Dict, List, Optional, Tuple
 
 from ...core.base_extractor import MetadataExtractor
+from ...core.positions import whole_sql
 from ...errors import ConversionRefused
 from ..types import ComprehensiveMetadata, EssentialMetadata
 
@@ -30,6 +31,19 @@ _CALIBRATION_INFO_TEXT = (
 )
 
 
+def _mismatch_reading(raster: Tuple[float, float], beam: Tuple[float, float]) -> str:
+    """What a Raster step that differs from BeamScanSize says, by direction.
+
+    Oversampled only when the beam scans more than one step, on some axis
+    and on none less; a step larger than the beam leaves gaps between spots.
+    """
+    if all(b >= r for r, b in zip(raster, beam)):
+        return "The beam scan is larger than the step: the acquisition is oversampled."
+    if all(b <= r for r, b in zip(raster, beam)):
+        return "The beam scan is smaller than the step: the spots leave gaps."
+    return "The beam scan is larger than the step on one axis and smaller on the other."
+
+
 def _float64_count(blob: Any) -> Optional[int]:
     """How many little-endian float64 values a ``CalibrationInfo`` blob holds."""
     if not isinstance(blob, (bytes, bytearray)) or len(blob) % 8:
@@ -298,26 +312,30 @@ class BrukerMetadataExtractor(MetadataExtractor):
 
         When a region is selected, only frames from that region are
         included in the bounds and count.
+
+        The bounds come from the frames whose position and region number
+        are whole numbers: the reader skips the others, and a text ``MAX`` ended in ``int()``
+        while a REAL just below a whole number was cut down (issue #527).
+        The count stays every frame, as it was.
         """
+        whole = " AND ".join(
+            whole_sql(column) for column in ("XIndexPos", "YIndexPos", "RegionNumber")
+        )
+        bounds = (
+            f"MIN(CASE WHEN {whole} THEN XIndexPos END), "
+            f"MAX(CASE WHEN {whole} THEN XIndexPos END), "
+            f"MIN(CASE WHEN {whole} THEN YIndexPos END), "
+            f"MAX(CASE WHEN {whole} THEN YIndexPos END), "
+            "COUNT(*) as frame_count"
+        )
         if self._region is not None:
-            frame_query = """
-            SELECT
-                MIN(XIndexPos), MAX(XIndexPos),
-                MIN(YIndexPos), MAX(YIndexPos),
-                COUNT(*) as frame_count
-            FROM MaldiFrameInfo
-            WHERE RegionNumber = ?
-            """
-            cursor.execute(frame_query, (self._region,))
+            where = " WHERE RegionNumber = ?"
+            cursor.execute(
+                f"SELECT {bounds} FROM MaldiFrameInfo{where}",  # nosec B608
+                (self._region,),
+            )
         else:
-            frame_query = """
-            SELECT
-                MIN(XIndexPos), MAX(XIndexPos),
-                MIN(YIndexPos), MAX(YIndexPos),
-                COUNT(*) as frame_count
-            FROM MaldiFrameInfo
-            """
-            cursor.execute(frame_query)
+            cursor.execute(f"SELECT {bounds} FROM MaldiFrameInfo")  # nosec B608
         return cursor.fetchone()
 
     def _query_frame_count(self, cursor) -> int:
@@ -581,6 +599,33 @@ class BrukerMetadataExtractor(MetadataExtractor):
         mis_path = find_mis_file_for_d_folder(self.data_path)
         return parse_mis_file(mis_path) if mis_path is not None else None
 
+    def _beam_scan_size(
+        self, beam_x: object, beam_y: object
+    ) -> Optional[Tuple[float, float]]:
+        """The beam scan size as two finite numbers above zero, or ``None``.
+
+        sqlite stores whatever was written, so a column declared REAL can
+        hold ``'20 um'``. A value that is not a number is treated like a
+        missing one, with a warning naming it, instead of ending the
+        conversion in ``float()`` even when ``--pixel-size`` was given.
+        """
+        if not beam_x or not beam_y:
+            return None
+        try:
+            beam = (float(beam_x), float(beam_y))  # type: ignore[arg-type]
+        except (TypeError, ValueError):
+            beam = (math.nan, math.nan)
+        if all(math.isfinite(v) and v > 0 for v in beam):
+            return beam
+        logger.warning(
+            "MaldiFrameLaserInfo BeamScanSize (%r, %r) in %s is not two numbers "
+            "above zero, so it is treated as missing",
+            beam_x,
+            beam_y,
+            self.data_path.name,
+        )
+        return None
+
     def _resolve_pixel_size_um(
         self,
         beam_x: Optional[float],
@@ -612,20 +657,19 @@ class BrukerMetadataExtractor(MetadataExtractor):
             ):
                 raster = (float(raster_xy[0]), float(raster_xy[1]))
 
-        beam: Optional[Tuple[float, float]] = (
-            (float(beam_x), float(beam_y)) if beam_x and beam_y else None
-        )
+        beam = self._beam_scan_size(beam_x, beam_y)
 
         if raster is not None and beam is not None:
             if raster != beam:
                 logger.warning(
                     "Pixel size mismatch: .mis Raster=(%g, %g) um, "
-                    "BeamScanSize=(%g, %g) um. Acquisition is oversampled. "
+                    "BeamScanSize=(%g, %g) um. %s "
                     "Using Raster step. Override with --pixel-size if needed.",
                     raster[0],
                     raster[1],
                     beam[0],
                     beam[1],
+                    _mismatch_reading(raster, beam),
                 )
             return raster