0.1.1b2: one type scale, measured layout, truecolor PDF rasters

CHANGELOG.md

@@ -3,10 +3,55 @@
 ## Unreleased
 
 ### cnplot package
+- Clone names are set in the regular weight: the clone labels beside the integer-CN profile
+  and the `<label> (<pct>%)` row-block labels on the heatmap. The CN key's `CNA` title is
+  regular as well.
 - `BAF_LIM = (-0.01, 1.01)`: default limits for a BAF-like axis on [0, 1], exported from
   `cnplot`. The pad keeps points at exactly 0 or 1 off the spine while the ticks still run
   0.0 to 1.0. Used by the gallery/README BAF row (`ylim`) and the RDR-vs-BAF scatter
   (`xlim`), which previously used `(-0.05, 1.05)` and `(0, 1)`.
+- Fonts are configured in one module, `cnplot.cnplot_style`: the family list
+  (`FONT_FAMILY`), the embedding types (`FONT_TYPE`), the `FS_*` sizes, and `apply_style()`,
+  which writes them into rcParams on import. No other module names a family, a size or an
+  embedding type. `FONT_FAMILY` and `apply_style` are exported from `cnplot`.
+- One type scale for every figure. The heatmap page, the integer-CN profile, the CN key,
+  the proportion legends and the 2D scatter no longer carry their own literal font sizes
+  (6 to 15 pt); they take `FS_TICK`, `FS_LEGEND`, `FS_AXLABEL`, `FS_AXTITLE` and `FS_TITLE`,
+  and the same sizes are set as rc defaults on import so text drawn without an explicit
+  size (a colorbar's ticks) follows. `FS_ANNOT` is new, for labels written on the data.
+- Layout measured instead of assumed, so it holds at any font size:
+  - the CN key labels each state `(a,b)` and wraps its groups onto further rows until the
+    widest label fits under one swatch; it never shrinks the type or the notation;
+  - `plot_strip_legend` stacks legends by their rendered height. Its `entry_h` argument is
+    removed and `gap` defaults to half a legend line;
+  - the heatmap colorbar grows until neighbouring tick labels clear each other, and the
+    strip legends start to the right of its rendered tick labels;
+  - the `plot_heatmap_cnp` title is anchored above the mesh rather than at a fixed
+    figure height.
+- Genome-axis labels are placed so they clear each other at the type scale:
+  - chromosome names keep their `rotation` (60 degrees by default); names closer together
+    than a line at that angle are nudged apart along the axis, and on an axes too narrow
+    to hold them in one row, alternate crowded names go on a second row (`chrname_tiers`
+    tells a layout which case applies);
+  - in Mb mode an Mb label within a line of the previous one is left blank (its tick mark
+    stays), the chromosome names under the Mb labels are offset in points rather than as a
+    fraction of the axes, and on a stacked page only the last row labels its Mb ticks;
+  - the gap between a row and the profile under it is sized in inches from the labels it
+    holds (`genome_label_room`) by `plot_scatter_1d_multisample` and `plot_heatmap_cnp`,
+    and `plot_heatmap_cnp` no longer opens the same gap between the profile and its key.
+- The README gallery is drawn 20 in wide, the width the type scale assumes.
+- `FigureSaver.close()` can be called twice on every supported matplotlib.
+- Importing cnplot stops matplotlib's PDF backend storing rasterized layers as `/Indexed`
+  palette images. With `pdf.compression` on, the backend converts any RGB raster of 256 or
+  fewer colours to a palette and packs the indices at 1, 2 or 4 bits per pixel when the
+  palette is small, which vector editors misread, so a copy-number raster opened in the
+  wrong colours. Rasters are now written at 8 bits per component in DeviceRGB, still
+  Flate-compressed. The hook is on `PdfFile._writeImg`, so it applies to every PDF the
+  process writes, through `FigureSaver`, `PdfPages` or `Figure.savefig`. Rendered pixels
+  are unchanged; a PDF holding such rasters is larger.
+  This works around a known matplotlib bug, wrong colours when such a PDF is opened in
+  Adobe Illustrator ([matplotlib#27893](https://github.com/matplotlib/matplotlib/issues/27893)),
+  for which matplotlib has no setting; remove the hook once that is fixed upstream.
 - `FigureSaver` no longer holds a whole-figure raster buffer per rasterized artist.
   matplotlib renders each rasterized artist into a buffer the size of the entire figure
   and passes the PDF backend a view of it, and the backend keeps every image until the

README.md

@@ -1,6 +1,6 @@
 # CN-Plot
 
-[![Version](https://img.shields.io/badge/version-0.1.1-blue.svg)](VERSION)
+[![Version](https://img.shields.io/badge/version-0.1.1b2-blue.svg)](VERSION)
 [![Python](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/downloads/)
 [![Code style: Ruff](https://img.shields.io/badge/code%20style-ruff-000000.svg)](https://github.com/astral-sh/ruff)
 
@@ -47,7 +47,7 @@ from cnplot import plot_cnv_profile
 
 seg_df = pd.read_table("sample.seg.ucn.tsv")
 
-fig, (ax, ax_leg) = plt.subplots(2, 1, figsize=(12, 3), height_ratios=[3, 1])
+fig, (ax, ax_leg) = plt.subplots(2, 1, figsize=(20, 3.5), height_ratios=[3, 1])
 plot_cnv_profile(ax, seg_df, genome_axis, sample_id="S1", ax_leg=ax_leg,
                  title="Integer copy-number profile")
 fig.savefig("profile.png", bbox_inches="tight")   # returns None; you own the figure
@@ -140,6 +140,7 @@ fig = plot_heatmap_cnp(
     title="Single-cell RDR heatmap",
     row_labels=cell_labels, cmap=cmap, norm=norm,
     cbar_label="RDR", cbar_ticks=ticks,
+    figsize=(20, 9),
     mb_ticks=True,                       # Mb ticks every 50Mb per chromosome
     strips=[
         {"name": "Copy-typing", "matrix": posteriors, "order": clones, "cmap": clone_cmap},

VERSION

@@ -1 +1 @@
-0.1.1
+0.1.1b2

examples/plot_gallery.py

@@ -28,6 +28,7 @@
 
 from cnplot import (  # noqa: E402
     BAF_LIM,
+    FS_TITLE,
     GenomeAxis,
     get_baf_cmap,
     get_log2rdr_cmap,
@@ -50,7 +51,7 @@ def save(fig, name):
 
 
 def suptitle(fig, text, y=1.0):
-    fig.suptitle(text, fontsize=14, fontweight="bold", y=y)
+    fig.suptitle(text, fontsize=FS_TITLE, fontweight="bold", y=y)
 
 
 genome_axis = GenomeAxis(*reference())
@@ -64,7 +65,7 @@ def suptitle(fig, text, y=1.0):
 )
 
 # 1. integer copy-number profile
-fig, (ax, lg) = plt.subplots(2, 1, figsize=(12, 3), height_ratios=[3, 1])
+fig, (ax, lg) = plt.subplots(2, 1, figsize=(20, 3.5), height_ratios=[3, 1])
 plot_cnv_profile(
     ax, sim.seg_ucn, genome_axis, sample_id="S1", ax_leg=lg, show_pi_viol=False
 )
@@ -166,6 +167,7 @@ def suptitle(fig, text, y=1.0):
         cbar_label=label,
         cbar_ticks=ticks,
         strips=strips,
+        figsize=(20, 9),
         mb_ticks=True,
         profile_kwargs={"show_pi_viol": False},
     )

src/cnplot/__init__.py

@@ -49,16 +49,21 @@
 from .cnplot_heatmap import plot_column_strips, plot_heatmap, plot_strip_legend
 from .cnplot_intcnp import plot_cnv_profile
 from .cnplot_io_utils import read_bed, read_chr_sizes
-from .cnplot_utils import (
-    AXIS_LABELS,
-    BAF_LIM,
-    CN_PREFIX,
-    EXP_PREFIX,
+from .cnplot_style import (
+    FONT_FAMILY,
+    FS_ANNOT,
     FS_AXLABEL,
     FS_AXTITLE,
     FS_LEGEND,
     FS_TICK,
     FS_TITLE,
+    apply_style,
+)
+from .cnplot_utils import (
+    AXIS_LABELS,
+    BAF_LIM,
+    CN_PREFIX,
+    EXP_PREFIX,
     MARKER_SIZE_LARGE,
     MARKER_SIZE_SMALL,
     MAX_NDOTS,
@@ -68,10 +73,12 @@
     U_PREFIX,
     FigureSaver,
     adaptive_dot_size,
+    chrname_tiers,
     decorate_genome_axis,
     display_label,
     draw_chr_boundaries,
     draw_segment_boundaries,
+    genome_label_room,
     get_clone_ylabels,
     get_transparency,
     resolve_colors,
@@ -104,6 +111,7 @@ def _read_version() -> str:
 __version__ = _read_version()
 
 __all__ = [
+    "apply_style",
     "AXIS_LABELS",
     "BAF_COLORS",
     "BAF_LIM",
@@ -112,6 +120,8 @@ def _read_version() -> str:
     "DATASET_CMAP",
     "DEFAULT_EXCLUDED_CHROMS",
     "EXP_PREFIX",
+    "FONT_FAMILY",
+    "FS_ANNOT",
     "FS_AXLABEL",
     "FS_AXTITLE",
     "FS_LEGEND",
@@ -138,7 +148,9 @@ def _read_version() -> str:
     "__version__",
     "adaptive_dot_size",
     "annotate_landmarks",
+    "chrname_tiers",
     "decorate_genome_axis",
+    "genome_label_room",
     "display_label",
     "draw_chr_boundaries",
     "draw_segment_boundaries",

src/cnplot/cnplot_1d.py

@@ -22,11 +22,13 @@
 from matplotlib.collections import LineCollection
 
 from .cnplot_genome_axis import GenomeAxis
-from .cnplot_utils import (
-    EXP_PREFIX,
+from .cnplot_style import (
     FS_AXLABEL,
     FS_AXTITLE,
     FS_LEGEND,
+)
+from .cnplot_utils import (
+    EXP_PREFIX,
     decorate_genome_axis,
     display_label,
     draw_chr_boundaries,

src/cnplot/cnplot_2d.py

@@ -28,6 +28,11 @@
     get_clone_states,
     select_sample,
 )
+from .cnplot_style import (
+    FS_ANNOT,
+    FS_AXTITLE,
+    FS_LEGEND,
+)
 from .cnplot_utils import (
     EXP_PREFIX,
     NORMAL_CLONE,
@@ -100,7 +105,7 @@ def annotate_landmarks(
     landmarks: list,
     markersize: float = 30,
     linewidth: float = 1.0,
-    fontsize: int = 9,
+    fontsize: int = FS_ANNOT,
     color="black",
 ) -> None:
     """Draw hollow markers with repelled ``(a, b)`` labels.
@@ -411,7 +416,7 @@ def plot_scatter_2d(
                 grid.ax_joint.legend(
                     handles=handles,
                     loc="best",
-                    fontsize="small",
+                    fontsize=FS_LEGEND,
                     fancybox=True,
                     framealpha=0.7,
                     handlelength=0,
@@ -421,9 +426,9 @@ def plot_scatter_2d(
     grid.set_axis_labels(
         xlabel=display_label(xcol if xlabel is None else xlabel),
         ylabel=display_label(ycol if ylabel is None else ylabel),
-        fontsize=12,
+        fontsize=FS_AXTITLE,
     )
     if title is not None:
-        grid.figure.suptitle(title, fontsize=12, fontweight="bold")
+        grid.figure.suptitle(title, fontsize=FS_AXTITLE, fontweight="bold")
     grid.figure.tight_layout()
     return grid

src/cnplot/cnplot_figures.py

@@ -29,16 +29,27 @@
     plot_cnv_profile,
     select_sample,
 )
-from .cnplot_utils import (
+from .cnplot_style import (
     FS_AXLABEL,
+    FS_LEGEND,
+    FS_TITLE,
+)
+from .cnplot_utils import (
     NORMAL_CLONE,
     SAMPLE_COL,
+    chrname_tiers,
     display_label,
+    genome_label_room,
     resolve_marker_size,
 )
 
 logger = logging.getLogger(__name__)
 
+# Share of a figure's width its main axes takes under matplotlib's default margins.
+_AXES_FRAC = 0.775
+# Space left between a profile and the key under it, in inches.
+_KEY_GAP_IN = 0.25
+
 __all__ = [
     "make_row_spec",
     "plot_heatmap_cnp",
@@ -89,7 +100,7 @@ def _draw_prop_legend(
         handles=handles,
         loc="center left",
         bbox_to_anchor=(1.01, 0.5),
-        fontsize="small",
+        fontsize=FS_LEGEND,
         fancybox=True,
         framealpha=0.7,
         handlelength=0,
@@ -263,14 +274,20 @@ def plot_scatter_1d_multisample(
     has_profile = seg_df is not None
     rows_h = row_height * n_rows * n_groups
     profile_h = 2.0 if has_profile else 0.0
-    fig_h = rows_h + profile_h + 0.5
+    # NB: the rows-to-profile gap holds the profile's names and the last row's Mb labels
+    tiers = chrname_tiers(genome_axis, _AXES_FRAC * row_width)
+    room = genome_label_room(mb_ticks=mb_ticks, tiers=tiers) if has_profile else 0.0
+    fig_h = rows_h + profile_h + 0.5 + room
     fig = plt.figure(figsize=(row_width, fig_h))
 
     # Nested so the three gaps stay independent: rows within a group, group to
     # group, and rows to profile. One flat grid would tie the last two together.
     # NB: a top-side xlabel sits in the rows-to-profile gap; 0.2 clears the row above.
     gap = max(profile_hspace, 0.2) if xlabel is not None else profile_hspace
     if has_profile:
+        # NB: hspace is a fraction of the mean cell height; solve it for `room` inches
+        avail = (0.97 - fig.subplotpars.bottom) * fig_h
+        gap = max(gap, 2.0 * room / max(avail - room, room))
         outer = GridSpec(
             2,
             1,
@@ -337,6 +354,9 @@ def plot_scatter_1d_multisample(
                 mb_tick_step=mb_tick_step,
                 rasterized=rasterized,
             )
+            if mb_ticks and not (last_row and last_group):
+                # NB: stacked rows share the x axis; only the last one labels it
+                ax.tick_params(axis="x", labelbottom=False)
             if last_row and show_props and seg_df is not None:
                 _draw_prop_legend(
                     ax, seg_df, group, group_col, normal, display_min_clone_prop
@@ -425,9 +445,13 @@ def plot_heatmap_cnp(
         raise ValueError("profile must be 'cnv' or 'ascn'")
     if profile_hspace is None:
         base = 0.45 if heatmap_kwargs.get("mb_ticks") else 0.35
-        # NB: hspace is a fraction of axes height; the labels are not, so short pages
-        # need a larger fraction to hold the same band.
-        profile_hspace = max(base, 4.5 / figsize[1])
+        # NB: hspace is a fraction of the mean axes height; solve it for inches
+        room = genome_label_room(
+            mb_ticks=bool(heatmap_kwargs.get("mb_ticks")),
+            tiers=chrname_tiers(genome_axis, _AXES_FRAC * figsize[0]),
+        )
+        avail = (top - plt.rcParams["figure.subplot.bottom"]) * figsize[1]
+        profile_hspace = max(base, 3.0 * room / max(avail - 2.0 * room, room))
     heatmap_kwargs.setdefault("show_colorbar", True)
     heatmap_kwargs.setdefault("show_block_labels", False)
     heatmap_kwargs["plot_chrname"] = False
@@ -439,6 +463,10 @@ def plot_heatmap_cnp(
         3, 1, figsize=figsize, gridspec_kw={"height_ratios": list(height_ratios)}
     )
     fig.subplots_adjust(top=top, hspace=profile_hspace)
+    # NB: hspace also opens the profile-to-key gap, which holds no labels; close it
+    prof, key = axes[1].get_position(), axes[2].get_position()
+    lift = max(prof.y0 - key.y1 - _KEY_GAP_IN / figsize[1], 0.0)
+    axes[2].set_position([key.x0, key.y0 + lift, key.width, key.height])
 
     plot_heatmap(axes[0], matrix, coords_df, genome_axis, **heatmap_kwargs)
     draw_profile(
@@ -450,5 +478,7 @@ def plot_heatmap_cnp(
         **profile_kwargs,
     )
     if title is not None:
-        fig.suptitle(title, fontsize=14, fontweight="bold", y=0.98)
+        # NB: anchored to the mesh, so a taller title grows the page upward
+        y = axes[0].get_position().y1 + 0.5 * FS_TITLE / 72.0 / figsize[1]
+        fig.suptitle(title, fontsize=FS_TITLE, fontweight="bold", y=y, va="bottom")
     return fig