thyra 2.0.2__tar.gz → 2.0.4__tar.gz

{thyra-2.0.2 → thyra-2.0.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: thyra
-Version: 2.0.2
+Version: 2.0.4
 Summary: A modern Python library for converting Mass Spectrometry Imaging (MSI) data into SpatialData/Zarr format - your portal to spatial omics
 License: MIT
 License-File: LICENSE

{thyra-2.0.2 → thyra-2.0.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "poetry.core.masonry.api"
 
 [tool.poetry]
 name = "thyra"
-version = "2.0.2"
+version = "2.0.4"
 description = "A modern Python library for converting Mass Spectrometry Imaging (MSI) data into SpatialData/Zarr format - your portal to spatial omics"
 authors = ["Theodoros Visvikis <t.visvikis@maastrichtuniversity.nl>"]
 maintainers = ["Theodoros Visvikis <t.visvikis@maastrichtuniversity.nl>"]

{thyra-2.0.2 → thyra-2.0.4}/thyra/__init__.py

@@ -36,7 +36,7 @@ warnings.filterwarnings(
     category=FutureWarning,
 )
 
-__version__ = "2.0.2"
+__version__ = "2.0.4"
 
 # Import key components - avoid wildcard imports
 try:

{thyra-2.0.2 → thyra-2.0.4}/thyra/converters/spatialdata/base_spatialdata_converter.py

@@ -1082,8 +1082,13 @@ class BaseSpatialDataConverter(BaseMSIConverter, ABC):
         # bincount without minlength is fastest
         accumulated = np.bincount(indices_clipped, weights=intensities)
 
-        # Extract non-zero bins for sparse representation
-        nonzero_mask = accumulated > 0
+        # Extract non-zero bins for sparse representation. The test is
+        # ``!= 0`` rather than ``> 0`` because a bin whose accumulated value is
+        # negative is still a measurement: dropping it silently raises the
+        # stored TIC above the input's. Baseline-subtracted data can carry
+        # negative intensities, though every export seen so far filters them
+        # out upstream.
+        nonzero_mask = accumulated != 0
         nonzero_indices = np.where(nonzero_mask)[0].astype(np.int_)
         nonzero_values = accumulated[nonzero_mask]
 

{thyra-2.0.2 → thyra-2.0.4}/thyra/converters/spatialdata/streaming_converter.py

@@ -129,10 +129,21 @@ class StreamingSpatialDataConverter(BaseSpatialDataConverter):
         n_x, n_y, n_z = metadata.dimensions
         n_pixels = n_x * n_y * n_z
 
-        # Estimate number of m/z bins after resampling using the same
-        # resolution path the axis builder will use, so the gate against
-        # PCS_SIZE_THRESHOLD_GB sees the real bin count.
-        if self._resampling_config:
+        # Prefer the axis that was actually built. ``convert()`` runs
+        # ``_initialize_conversion()`` -- and so ``_setup_mass_axis()`` --
+        # before it asks which method to use, so by the time this gate is
+        # evaluated the real bin count is known and there is nothing to
+        # estimate. That matters most on the raw-axis path, where the
+        # fallback below assumes a 10 mDa spacing that the data need not
+        # have: a continuous file carrying 4,000 points over 250-1200 m/z
+        # was scored as though it had 95,000, and a processed file whose
+        # spectra share no m/z values was scored far too low, which routed
+        # the largest datasets to the method that holds the most in memory.
+        if self._common_mass_axis is not None:
+            n_mz_bins = len(self._common_mass_axis)
+        elif self._resampling_config:
+            # Same resolution path the axis builder will use, so the gate
+            # against PCS_SIZE_THRESHOLD_GB sees the real bin count.
             _, _, _, n_mz_bins = self._resolve_resampling_plan()
         else:
             min_mass, max_mass = metadata.mass_range

{thyra-2.0.2 → thyra-2.0.4}/thyra/resampling/strategies/nearest_neighbor.py

@@ -2,6 +2,23 @@
 
 This strategy is optimal for centroid data (e.g., from timsTOF
 instruments) where each peak represents a discrete mass value.
+
+Not to be confused with the converters' ``_nearest_neighbor_resample``
+(``thyra/converters/spatialdata/base_spatialdata_converter.py``), which is
+what ``ResamplingMethod.NEAREST_NEIGHBOR`` actually selects during a
+conversion. The two share a name and do opposite things with intensity:
+
+- This class INTERPOLATES. Each target point takes the intensity of the
+  nearest source point, so one source peak is replicated across every
+  target point closer to it than to any other. Summed intensity therefore
+  scales with how densely the target axis is laid out, and is not
+  preserved. That is the defining behaviour of nearest-neighbour
+  interpolation, not a defect -- but it means this class is the wrong tool
+  if you care about total ion current.
+- The converter's version BINS. Each source peak is accumulated into
+  exactly one target bin, so summed intensity is preserved exactly.
+
+If you want the conversion pipeline's behaviour, use the converter.
 """
 
 from typing import Any
@@ -24,6 +41,10 @@ class NearestNeighborStrategy(ResamplingStrategy):
         and assigns its intensity. This preserves the discrete nature
         of centroid data.
 
+        Does NOT preserve total ion current -- a source peak is repeated
+        at every target point nearest to it, so the sum grows with the
+        target axis density. See the module docstring.
+
         Parameters
         ----------
         spectrum : Spectrum

{thyra-2.0.2 → thyra-2.0.4}/thyra/resampling/types.py

@@ -70,7 +70,13 @@ class MassAxis:
         return np.diff(self.mz_values)
 
     def resolution_at(self, mz: float) -> float:
-        """Calculate resolution at given m/z."""
+        """Return ``mz`` divided by the local bin width.
+
+        This is a property of the AXIS, not of the data on it. A finely
+        spaced axis reports a large number here whether or not the spectra
+        stored on it resolve anything at that scale, so it must not be
+        reported to users as the acquisition's resolving power.
+        """
         idx = int(np.searchsorted(self.mz_values, mz))
         if idx > 0 and idx < len(self.mz_values):
             delta_mz = float(self.mz_values[idx] - self.mz_values[idx - 1])
@@ -91,8 +97,11 @@ class ResamplingConfig:
         axis_type: Mass axis spacing model.  ``None`` auto-detects from
             the instrument metadata.
         target_bins: Number of bins in the resampled axis.  ``None``
-            lets the resampler choose a bin count that preserves the
-            native resolution.
+            derives the count from ``mass_width_da`` at ``reference_mz``
+            across the mass range.  Note that this is a fixed target
+            width, not a measurement: the source spectra's own point
+            spacing is never consulted, so the resulting axis can be
+            finer or coarser than the data it will hold.
         mass_width_da: Bin width in Daltons at ``reference_mz``.
             Alternative to ``target_bins`` -- specify one or the other.
         reference_mz: Reference m/z for ``mass_width_da``.  Default