venndir_todo.md

In jmw86069/venndir: Directional Venn diagrams

todo on venndir

01may2023

Note the official TODO was moved to TODO.md.

03apr2023

• check build issues on R-4.2.3, dependency on data.table

06feb2023

• add venndir_legend() to summarize number of elements per Venn set.

• venndir()

• consider alternative label layouts:

  • position "overlap title" at top-center
  • one column underneath when overlap_type="overlap" with counts
  • two columns underneath, total counts on left, signed counts on right
  • or one column with counts, then signed counts in the same column

• output really needs to be its own object type

  • convert from list to something like class="venndir_output" which inherits from list, or is an S4 object with proper methods.
  • input to functions like nudge_venndir_label() should take a proper object instead of list.

• make it clear how to display overlap titles, e.g. set="all" is not working. The x$label_df[,"overlap"] column accepts "outside", "inside", or "none", and is not being filled correctly.

• plot_warning=TRUE when the label contains more than 3 overlap sets, it should just indicate the number of overlap sets rather than list each overlap set.

• DONE. scale default padding value based upon font_cex

• DONE. currently padding="4 pt" is too much when the font is shrunk down small

• DONE. padding should be an argument to venndir()

• add option for custom label placement?

• this option might not be needed if label orientation can be customized, e.g. with title centered above all count labels underneath it, instead of being offset.

• center by label box

• center by coordinate zero (the point between left/right columns in labels)

• nudge_venndir_label()

• add proper function parameter documentation

• x_offset, y_offset should be additive to existing values
• should be possible to set a label "inside", "outside"

• should be possible to set the overlap title "inside", "outside".

• error with label_style="shaded" from colorjam::blend_colors() likely caused by blending NA or NULL colors. See colorjam-blendcolors.R line 161.

• store more data inside venndir() output object

• Main goal is for render_vendir() to be totally self-sufficient when operating on venndir() output.

• Essentially all new parameters used in render_venndir() should be stored in the object, for example: plot_warning

• label segments

• the segment buffer should use a scale relative to the plot scale (e.g. 1% of the overall plot layout range), however when the polygon cannot accommodate that buffer, it should use progressively smaller scale relative to that polygon.

• outside label positions

• ideally during rendering, labels should be "dodged" in some way to minimize overlaps.

20oct2022

• venndir() possible bug

• when displaying proportional=TRUE, overlap_type="overlap" it does not indicate when a non-overlap cannot be displayed, for example items unique to one set. Found with 4-way Euler diagram.

• Appears to happen sometimes when eulerr returns circular instead of ellipse shape, which seems to happen due to internal randomness.

• min_count should be stored in mem output

• mem_enrichment_heatmap() uses min_count with p_cutoff to define significant pathways

• mem_plot_folio has no argument min_count and cannot get it from mem

• multiEnrichMap() should store min_count in the output mem

• new S4 classes

• mem: required for Bioconductor submission (if pursued), and useful otherwise

• mpf: or mem_plots or mem_folio with output from mem_plot_folio()

• mem_enrichment_heatmap()

• apparent bug when trying to order rows to match the order in mem_gene_path_heatmap() - but a "bug" because it is unsupported, but it does cause incorrect heatmaps when using ComplexHeatmap::draw(hm[row_index,]).

• When supplied with row_order this function should properly order by names(row_order) to match the actual rownames(enrichIM) used for the heatmap.

• argument min_count is not explicitly passed by mem_plot_folio() and not stored in the mem object.

• Keep track of this URL for installation issues in Mac OSX regarding sf: https://github.com/r-spatial/sf/issues/1536#issuecomment-727342736

20sep2022

• overlap_type="agreement" should hide agreement for single-set overlap.

• when overlap and signed overlap are displayed together, the label should center the main overlap number, with signed labels off to the side.

• Currently, main overlap and signed overlap are combined into one box, the box is centered on the label coordinate. However, inside a polygon sometimes the main overlap number is offset outside the polygon, and the signed overlap numbers are also offset, so really no numeric labels are positioned well. By default the main overlap number should be inside the polygon.

• Design idea: nudge label by overlap name.

• The idea is to make it easy to adjust a label.

• Something like this: R nudge_label=list( `A&B`=c(x=1, y=0), `A&B&C`=c(x=0.5, y=0), `B`=c(x=0, y=1))

01sep2022

• Small bug fix:

• overlap_type="overlap",

• label_preset="main items",
• show_items="sign item"
• It displays the counts instead of the sign, since there is no sign when overlap_type="overlap".

• when overlap_type="overlap", the show_items argument should ignore "sign" as a possible display value.

• signed_overlaps() should default to non-signed overlaps when all entries in setlist have incidence value 1, "up".

• when setlist is an overlap list, convert with overlaplist2setlist() to setlist.

• Conversion functions, and inputs should be more user-friendly:

• signed_overlaps() argument setlist should accept output from label_df$items.

• counts2setlist() and signed_counts2setlist() should accept the names from lengths(label_df$items), which includes a suffix with the overlap hit values.

  • For example, names include: "GroupA|1 0", "GroupB|0 1", "GroupA&GroupB|1 1". Simple enough, they should be equivalent to: "GroupA", "GroupB", "GroupA&GroupB".
  • Also, for signed overlaps, the suffix indicates the direction, and should be honored.

31may2022

• Bug with textvenn() caused by jamba::printDebug().

23may2022

• Bug: proportional=TRUE and show_items="sign item" appears to cause it to position labels inside a polygon that does not exist. See venndir.R line 817 inside render_venndir(). Error: "Error in label_xy[, 1] : subscript out of bounds" seems like a symptom of a polygon that does not exist in proportional polygons.
• Why is there such a large gap between overlap and signed counts in the label? See if there is a padding that can be reduced in the respective ggtext labels.
• Consider new default: overlap_type="each".
• Consider new default: label_preset="main outside".
• Consider new default: inside_label_percent=0 or 1.
• Consider by default positioning labels above/below rather than radially around the plot, otherwise consider expand_fraction=0.2.

• inside_label_percent should use the bounding box area as the basis for the calculation, since the 4-way Venn takes up a larger percentage of the bounding box than the 3-way Venn. The theory should be the percentage of screen space available to position a label.

• Actually, "ideal world" the theory should be whether the label fits inside the Venn overlap polygon. The label bounding box will generally be larger than the actual label text, so this approach is more likely to determine labels are "too big" when they could actually fit inside the polygon.

• Single numerical counts are most intuitive inside the polygon.
• Signed counts are "busy" - but when they fit beside the main overlap count without overstepping the polygon, it works well.

15may2022

• scale proportional coordinates to rough the same magnitude as non-proportional coordinate circles.

• proportional coordinates tend to have much larger (7x) magnitude

• other downstream components are based upon absolute coordinates:
• item_buffer, segment_buffer
• offset

06may2022

• adjust item label size item_cex automatically based upon

• number of item labels

• relative area of the polygon

• venndir() and render_venndir() should have specific argument for the item label buffer width

• it currently uses label_polygon_fill() argument scale_width which is impossible to remember!

• Prepare some easy way to enable Venn diagrams from sestats results from jamses::se_contrast_stats().

• Consider something similar for limma::decideTests() which returns an incidence matrix.

• Consider allowing incidence matrix as valid input to venndir().
• While we're at it, try to recognize other common inputs, such as signed count vector, count vector, incidence matrix, signed incidence matrix.

Convert from SpatialPolygonsDataFrame to sf equivalent

Background:

• The sp package is deprecated, to be retired by end of 2023.
• All sp methods must be converted to use sf format.
• Fortunately, sf uses something like a data.frame as its storage model for polygons. Most if not all functions from rgeos are already available as sf-specific functions, the steps just all need to be ported.

• Unsure if spsample is available for sf. It would not be too difficult to wrapper a very simple replacement function:

• start with square coordinate "grid"

• rotate grid 30, 45, or 60 degrees
• overlay grid onto polygon P
• resize grid progressively smaller until at least N points are inside P
• optionally perform basic rotations at each sizing to maximize fit

28apr2022

• render_venndir() does not honor show_segments=FALSE with plot_style="gg".

• In the rare case that two sets completely overlap, proportional Euler returns two identical circles. We should probably offset them at least 1% so the circle colors are visible.

• venndir() obtains the proportional coordinates, get_venn_shapes() creates the coordinates - the step that creates coordinates should probably test and do the adjustment, so any function that calls get_venn_shapes() will receive valid coordinates.

• Problem: proportional Venn labels outside are not placed in ideal locations:

• often placed left and right side, which extend beyond the panel width

• sometimes still overlap other labels placed outside, partly due to irregular label "box", wide text labels, multi-line text labels, multi-component labels (set name, counts, signed counts)

• Need a method to display overlap count and item labels in a set.

• When item labels exceed max_items it should display the overlap counts, currently it displays nothing.

• Need a method to display set sizes somewhere in the plot panel:

• Total items in each set.

• Total items overall.

• Need method to display overlap percent as alternative to overlap count

• How do other Venn diagram tools calculate percent overlap?

09mar2022

• textvenn() is useful but sometimes labels are super long and fill the screen width (and beyond). It needs some logic to word-wrap. E.g. wrap by the separator "&" as first-pass; potentially word-wrap each label within that if longer than a defined length. Word-wrap probably needs to be sensitive to underscores "_", hyphens "-", period ".", and whitespace " ", including things like "\t\r\n", etc.

02feb2022

• Option to supply specific SpatialPolygons objects for each set.
• Option to apply rotate_degrees to the polygon representing specific sets. The function rescale_sp() takes rotate_degrees as an argument then applies to all polygons. It could potentially be applied to named polygons, if rotate_degrees also had names that matched the sp object.

30jan2022

Bugs observed

• Sometimes label_style="shaded box" throw error and do not display.
• Sometimes label_preset="main outside" causes main labels not to be plotted. The inside labels appear correctly.
• Proportional Venn labels draw a segment to an ambiguous polygon, sometimes two segments end inside the same polygon.

Feature enhancements

• venndir() option to "recognize" when the input is format:

• incidence matrix: im2list()

• signed incidence matrix: im_value2list()
• overlap counts: counts2setlist(); force return_items=FALSE

• signed overlap counts: signed_counts2setlist(); force return_items=FALSE

• Option to supply a list where some elements are signed, while others are not. For example, supplying "dysregulated genes" with direction, and supplying "genes assayed" which is not directional. This feature implies certain downstream effects:

• Concordance cannot be confirmed or denied with non-directional list elements. For example, "gene hit down", "protein hit down", "protein assayed" should be reported as "concordant", and the label should be something like "downArrow hyphen downArrow".

• Need more/better styles for arranging labels.

• Option to center set name above the counts and signed counts.

• Option for multi-column signed counts, especially for 3+ signed sets.

15nov2021

• indicate total items per set somewhere in the figure

• different label layout strategies, for example:

• The main set label is placed top-left, main counts bottom-left, signed counts placed top-right through bottom-right. The label width is calculated and centered inside the Venn polygon, however the counts are sometimes pushed to the far right if the main set label is too long.

• Alternative is to place set label top-center, then main count labels bottom-left, signed counts bottom-right.

02nov2021

• proportional Venn label positioning bugs:

• when a circle is fully inside another, it tries to label the interior circle using the polygon with fewest overlaps; however it errantly tries to place the count label for the unique set also inside the polygon which is obviously incorrect.

• it would be good to re-draw Venn circles using the actual set color after everything is rendered - to help reinforce the set colors

• "Try harder to make Euler proportional Venn diagram layout that represents all overlaps." A related issue was filed until R "eulerr" Github here: https://github.com/jolars/eulerr/issues/84

• basic idea is to manipulate the algorithm eulerr/src/optim_final.cpp lines 125 to 131. Example code is shown that adjusts the scoring.

• The current approach ranks solutions by lowest residual of fitted_area - orig_area - in other words absolute area represented. My thought was to add an "area penalty" when an entire overlap is not represented - this penalty could be a percentage of the total counts that are not represented multiplied by a weighting factor, for example 5x.

29oct2021

• Lot of examples of Venn diagram includes the percentage either along with the overlap count, or instead of the overlap count.
• Several examples of Venn diagrams hide/display a subset of labels of interest.

• Lot of venndir examples where the set label is super long, which makes the center of the label offset in a weird way.

• ideal world: have some configurable orientations of labels to include in each cell.

• one example: set title is centered across the label, the overlap and signed overlaps all appear below this title.

01oct2021

• Make curate_venn_labels() options more clearly defined

• the default data should be documented somewhere, perhaps as data to make it easier to customize

28sep2021

• I cannot ever remember how to add item labels! If I cannot remember, nobody else can remember. It needs to be easy to enable.

• Some option to word wrap super long set labels, otherwise the label is 10x wider than the plot size.

• COMPLETE: Implement max_items so item labels are hidden in sections with too many labels.

• COMPLETE: Item labels should use text() or geom_text() and not the gridtext() and geom_gridtext() counterparts - they are extremely slow for even moderate (20 to 50) labels on one figure. The gridtext feature is mostly useful for markdown support, which can be optional.

• It is hard to adjust placement of labels outside the circle.

• offset is intended to help push labels in one direction, but does not work reliably, especially for proportional Euler diagrams. One option is to scale proportional coordinates to typical dimensions used by fixed layouts.

• Revisit the workflow to nudge or move label positions.

• COMPLETE: polygon_label_fill() argument apply_n_scale=TRUE should also adjust polygons by relative size of the polygon compared to the size of the overall Venn diagram. Smaller polygons with fewer labels should not be adjusted as much as larger polygons.

• In general, proportional diagrams should use the same buffer for each polygon, since the polygon is already proportional to the number of item labels.

• polygon_label_outside() should adjust relative buffer consistent with polygon_label_fill(), based upon the plot size.

• More arguments should be added to venndir() so these options are not "hidden" features:

• shape="ellipse" is used for proportional Euler diagrams and allows elliptical ovals instead of circles.

• scale_width is passed to polygon_label_fill() when items are being displayed.
• draw_buffer - shows item polygon buffer
• segment_buffer - adjusts segment from outside labels to respective internal polygon.
• apply_n_scale - adjusts item polygon buffer by number of labels.

• center_method, vector_method, segment_method passed to polygon_label_outside() to position labels around the Venn polygons.

• Longer term, I might need to adjust the "optimization settings" used by the eulerr package. It appears to perform a solver minimization function that maximizes the total overlap count represented. I think it could be penalized when a set overlap is not represented to "encourage" it not to skip a potential overlap.

Venn section labeling bugs

• Problem: sometimes an outside label points to wrong polygon inside the Venn diagram. Seems limited to proportional Euler diagrams.

• It appears the priority in selecting a polygon is sorted wrong, for example, it should pick the subsection with fewest overlapping sets. It seems to choose the whole Venn circle instead of a subsection.

• A main set label outside may have a segment that points to some arbitrary location.

• Sometimes an overlap count appears in a random location, and it appears to be "buggy". Sometimes the overlap count should be hidden; other times the overlap count is in the wrong location.

• Change render_venndir() to draw each Venn set border after the individual polygons have been drawn, but before the labels have been added. The goal is for the Venn set borders to match set_colors.

• Currently when a Venn circle is fully inside another circle, the two set colors are blended to produce the polygon fill and polygon border. In this case, the border does not match any value in set_colors.

25jun2021

• when input character vectors happen to have names, signed_overlaps() should not assume directional overlap unless specifically requested, otherwise it takes ages! Workaround is to use overlap_type="overlap".

usability of custom changes

• document using offset so outside labels can be influenced up or down, left or right in a figure.

low usability for item labeling

• It's too confusing to add item labels to the diagram, it needs to be more "automatic" so that when show_items is enabled, label_preset is updated consistently. Also label_preset needs to be visible in venndir() and render_venndir() instead of being passed via ....
• Item labels using ggtext and gridtext are PAINFULLY slow to render more than 100 or so labels. Slow in terms of minutes, compared to the expected sub-second response time. Maybe the developers are okay with that, but there is no big reason not to use something orders of magnitude faster in venndir... like text() and geom_text(). It means item labels cannot contain markdown, unless we allow ggtext and gridtext as an option, like for venn_meme() when there is usually only one item per overlap. Using text() and geom_text() may impact the font and Unicode compatibility of displaying arrows (again).

new label preset

• set label and count inside, signed labels outside

usability of custom label placement

• Use case: move one overlap label inside, another label outside the polygon.
• Use case: move the signed labels for one overlap outside, but keep set and main count label inside.

bugs

• When set label is supposed to be inside, but there is no portion of the circle unique to that set label, the label seems to go near a border. However, it often overlaps the count label, and maybe should be grouped with the largest internal polygon overlap for that set.
• Item labels ignores the argument max_items and should instead hide item labels and display numeric counts when the number of items is too high.
• Sometimes with 4-way proportional Venn diagrams the main set label seems to point to the whole circle, instead of pointing to the subset of the circle that is specific to the label. Something seems to be going wrong when choosing the best available polygon to direct the label. It appears to work for intersect polygons, but not for main set labels, which suggests the order of choosing a polygon when more than one has a matching label is the problem -- and this should only happen with main labels because they are associated with the full circle, and with any overlap polygon that only involves that same label.

• Item label buffer_w and buffer_h appear to use absolute plot units instead of relative units. Since proportional coordinates tends to be different range than the fixed Venn circles, the units are not interchangable, making it confusing. Also, these buffers should be apparent in venndir() and render_venndir() rather than being hidden inside ....

• COMPLETE: If input contains factor values, they are coerced to integer which is incorrect. It needs to convert to character then follow the proper steps.

vignette on fonts

From user testing and feedback, there are issues around font display. For example, pdf() output sometimes displays no text labels, while png() does display the text labels. We think this issue is due to the fontfamily argument and/or family parameter referencing a font that is not present on the user environment.

R fonts are sort of a nightmare. There are fonts for several different output devices, and they are not always consistent.

Choosing a font that does not represent Unicode characters will result in directional arrows being shown as blank squares, something like []. This issue is usually caused by one or two issues:

• Font does not include Unicode symbols at all.
• R locale does not permit Unicode characters.

Both the above must be valid for Unicode arrows to be displayed.

The vignette should describe several different scenarios and potential workarounds. Include example with Cairo package and CairoFonts().

enhancements

visual issue

Currently the main count label font is always the same color as the main set label font, even when the main set label is outside and the main count label is inside the polygon. Visually this is only a problem when the polygon fill is a dark color, and the outside background is a light color (usually white). When this happens, the count label inside the polygon is black on a dark color, and is very difficult to read.

The best current workaround is to use poly_alpha=0.2 which should force the Venn fill to be pale enough that black labels are clearly visible.

A refactor is required to store the main set name label color separate from the main count label. This change can be done two ways:

1. adding a row in label_df, which will also allow specifying the set name fill and box independent from the count fill and box.
2. adding a column to venn_spdf for font color, note this step does not allow the outside set name label to have different fill and box outline than the inside count label, if desired.

Looks like option 1 is preferable.

bugs

• render_venndir() argument max_items is being ignored. For overlaps with too many items, desired behavior is to display the counts instead of item labels.
• polygon_label_outside() is glitchy for weird proportional cases, for example 4 sets and shape="ellipse" might warrant different default settings. Alternative method would involve taking boundary point, expanding the boundary, find nearest point on that boundary, then proceeding with spread_degrees().

performance

• Showing a few hundred item labels is still painfully slow. Consider text() as an alternative. Or test options that remove the dither options for color/degrees/cex in case that somehow triggers non-vectorized rendering.
• polygon_label_outside() for 5-way Venn overlaps sometimes moves labels to weird positions. Workaround with min_degrees=8 but not sure why.

miscellaneous

• Option to display concordance score below the signed values.
• For overlap_type="agreement" hide the signed values for single-set overlaps, they always "agree" and are not informative.

prepare for R-shiny

• make ggtext::geom_textbox() and ggtext::geom_richtext() work for plotly::ggplotly(), by adding to_basic.*() functions. For example, in the package splicejam to make ggforce::geom_shape() work, I added the function to_basic.GeomShape().

• to_basic.GeomRichText()

• to_basic.GeomTextBox()

fixed bugs version 0.0.11.9000

• FIXED: im2list() with logical matrix input was incorrectly assigning all values. The empty argument will match the class of the input x to account for this mismatch.
• FIXED: show_zero=TRUE is not working properly, it is always hidden.
• FIXED: alpha_by_counts appears to be mis-aligned.

• FIXED: Fix issue where directional label colors apply make_color_contrast() without using the correct alpha value. See with venndir(make_venn_test(100, 3), poly_alpha=0)

• DONE: Add warning message to ggplot2 output in render_venndir().

vignettes

interfacing with VennDetail

• follow VennDetail vignette, show alternative with directional changes

interfacing with DESeq2/edgeR

add vignette showing more visual options

• how to change Venn set colors
• how to nudge a Venn circle
• how to rotate the whole Venn diagram
• how to adjust item label buffer, and view the buffer size
• how to create custom curations with curate_df
• how to handle gene data in a dataset with multiple probes per gene

make displaying items labels easier

• make it "one step" to display items, when show_items is enabled in venndir() the default label_preset should include items.
• Option to allow "one column" of labels inside the polygon, which could be useful for things like a list of pathways between two experiments.

polish label coordinates inside/outside

• coordinates for inside/outside should really just be in venn_spdf then used by label_df when needed
• label styles need polishing. If overlap label is outside and count label is inside, how can they have two different styles?

Combine individual labels into one cohesive label

COMPLETE for base R plots. TODO: for any label associated with a line segment, use degree_to_adj() to help place the label at the correct edge of the line segment. one option is simply to create one label, but text alignment for two column output would be difficult. * prototype using one label for fill/border, and expand the buffer size so it includes the other labels.

• calculate label width/height (grid::ascentDetails(), grid::widthDetails()
• also use grid graphics logic to use rescalable operations
• requires knowledge about grid graphics I don't currently have and was not yet easy to grok from online guides. I was able to proof-of-concept a method that used grob dimensions of a left label, to produce a right label with border and padding sufficient to encompass both sides. When trying to automate, it stopped working. Something in the circular logic, that a grob does not have a size until placed, so adjusting by size before placement has no usable value?

DONE: Update Rmarkdown README.Rmd and venndir_gene_expression.Rmd

• Use new recommended workflow and function arguments.
• Beware errors in venndir_gene_expression.Rmd with unique errors related to ragg_png, see details and workaround suggested here: https://www.jumpingrivers.com/blog/r-knitr-markdown-png-pdf-graphics/

DONE: small features for version 0.0.9.9000

• DONE: Some importer for signed overlaps. Given a known directional Venn, have a method to import to use venndir.

  • signed_counts2setlist()

• DONE: Bug: Fix polylabelr::poi() when there is one or more holes inside a polygon.

• DONE: change all arguments with angle to use degrees for consistency, and to make clear the type of angle.

• DONE: Refactor to handle "set name" separate from "main count" in the count label, so the "set name" can be moved, for example to perimeter or outside the polygon, leaving the numeric count inside as needed.

• DONE: Proportional diagram with one circle fully inside another does not have set name displayed -- it must be. Consider choosing one internal polygon with fewest set overlaps with the largest area.

• DONE: need idea for where to store the set label coordinates, since the Venn circle/ellipse shape does not have its own row.

• DONE: maybe venn shapes need to be stored in another list element returned by venndir()? It makes that output a bit too heavy imo.

• Maybe: new function venndir_bender() to modify a venndir result. resize, move, rotate venndir circles/ellipses.

• DONE: change default to return_items=TRUE for venndir() and signed_overlaps().

DONE: refactor label position

Design idea: Four label types: set, overlap, count, items

• set

• inside

• outside
• none

• ifhidden

• overlap

• none

• inside

• outside

• count

• inside

• outside

• none

• items

• none

• inside

Data storage:

• venn_spdf

• contains each Venn circle

• contains each Venn overlap polygon

• label_df

• contains each overlap count label

Progress:

• COMPLETE for base R plots.
• COMPLETE for ggplot.

DONE: update polygon_label_outside() for multiple labels

• DONE: The idea is to position labels by angle relative to the same central point, detect labels "too close" to one another, then spread them out evenly to assist with spacing between labels.
• Labels would be defined mainly by center point and angle in degrees.
• "Too close" would be defined by user-configurable minimum degrees, with some suitable default value.
• return the point of label, and the point at the polygon so a line segment can be drawn
• return the text adjust (adj) value so a text label can be placed at an appropriate side relative to the incoming line segment.

features for wider roll-out

make render_venndir() work for ggplot2 output

• Essentially roll ggrender_venndir() into render_venndir() then remove ggrender_venndir() (or make it a silent alias.)
• Add missing counts text label to ggplot2 output. Plan to use ggtext::geom_textbox() for a wider label in ggplot2.

Miscellaneous items

option to place set names at periphery of Venn circles

• This change could be most relevant with proportional Venn diagrams especially when one circle is fully enclosed inside another circle.
• outer label placement will be necessary when displaying items inside each polygon. Even if I could determine the count label size inside the polygon to place labels around it, it seems non-ideal
• this process is different for proportional circles which may be placed at various locations
• for Venn circles and ellipses the coordinates could be pre-defined

improve item label efficiency

• large number of labels makes it very inefficient -- find a way to make it faster.
• one idea: instead of calling gridtext for each polygon, aggregate them into a data.frame then display them in one call. Doubtful this change will be much faster.
• another idea: the "sp" spsample() function success rate is fairly low, requiring iterating with higher n + i labels in order to get n coordinates inside the polygon area. This iteration might be rate-limiting with large n.

make output compatible with plotly (or something interactive)

• Currently ggtext::geom_richtext() is not compatible with plotly::ggplotly() but probably just lacks the to_basic wrapper function. (See "splicejam" to_basic.GeomShape() for an example workaround.)
• This feature may require text labels at 0 or 90 degrees.

venn meme function

• basic idea is to make it "easy" to create a Venn meme when needed. The ability is already in place, but seems to have very consistently different default values than other Venn styles.

example of displaying items in large label box beside the Venn diagram

• manual placement of a larger label block where it handles word-wrapping itself, done by gridtext.

display fully "closed" Venn counts

• "Closed" includes the universe of each input list which is not represented in the Venn circles. For example:

• genes tested in set_A that are not hits

• genes tested in set_B that are not hits

• This option is based upon limma package display of total genes in each set, displayed outside the main Venn circles.

item labels

• It still sometimes fails to find enough labels inside the polygon, usually for large number of labels. Workaround might be for the user to change layout type to something like "regular" or "nonaligned" which could improve the search success rate.

• Add example to help text, and to Rmarkdown vignette.

• DONE: When there is only one item, consider using the center label instead of choosing a location inside the polygon, which sometimes chooses something away from the center.

• Note that sometimes the polylabelr::poi() point is outside the polygon, which happens sometimes when the polygon has a hole or otherwise "weird" shape. The method checks for that and uses the buffer polygon spsample() approach if the polylabelr::poi() point is not contained in the source polygon.

label_polygon_fill() with relative buffer scale_width

• add option for scale_width to be relative to the object size instead of absolute dimensions. For example:

• scale_width=-1 would produce no output polygon, because -1 is interpreted -100%,

• scale_width=-0.99 would produce an extremely small polygon, roughly 1% the original width,

• scale_width=-0.5 would produce a polygon with roughly half width buffer.

• The test example should use a relatively tall-thin ellipse, rotated to 45 degrees. The bounding box would be quite large, but the actual width of the polygon would be relatively small.

label_polygon_fill() extra features

• DONE: dither_cex,dither_color,dither_angle to adjust each text label slightly in three ways: cex to make label font slightly larger or smaller; color to make label color slightly lighter or darker; angle to adjust the angle of text slightly more or less than the original angle value. These effects can help visibly distinguish large number of labels packed into a relatively small polygon area.

missing counts warning

• DONE: The warning needs to check both show_label and show_items for each overlap set before deciding that a set is not displayed.

consider refactoring "base R plot" to use "grid graphics"

• Lower priority evaluation.
• Requires learning grid graphics in much more detail.
• The main benefit would be compatibility with gridtext::richtext_grob() which uses grid graphics. Currently using gridBase to combine grid and base R plotting works in interactive R sessions, but in Rmarkdown it is sometimes glitchy, because Rmarkdown now calls ragg::agg_png() and I'm almost sure this use would not be recommended or supported. "Pick one method" is a reasonable thing to ask of me.
• Another benefit is potential to calculate label width in a way to detect overlaps, and/or resize font size to fit relative to other labels.
• Another possibility is to re-position labels using relative units, for example move main/signed labels to the left by 1.2x main label width.

refactor venn count label styling

• DONE: venndir_label_style()

• currently resides inside venndir() which makes it hard to change the style in render_venndir(). I guess ideally the count label style can be updated by a separate function to make it easy to change styles.

calculate concordance coefficient

• Kruskal concordance: (agree - disagree) / (agree + disagree)
• Return concordance for each overlap set

Completed items

DONE: option to move count labels outside the polygon

• DONE: implement x_offset,y_offset for count labels

• Draw a line to polygon border if needed (polygon_label_segment())

DONE: refactor venndir output

The changes below are DONE.

• each polygon is represented per row in venn_spdf

• label (set overlap)

• venn_counts
• count label x,y coordinates
• count label x_offset,y_offset to displace a label outside the polygon (actually the offset will be stored in label_df for each label)
• polygon fill, border

• alpha, lwd, lty

• each label is represented in label_df, sometimes multiple labels per polygon

• venn_counts

• text (visible label)
• overlap_set (matches venn_spdf$label)
• type (main for total counts, signed for directional counts)
• x,y coordinates for each label (they should move together but not required)
• x_offset,y_offset for each label (make sure to draw each unique line only once)
• col (label text color), fill (label box fill), border (label box border)
• fontsize, vjust, hjust, halign, rot
• show_label - whether to display text for this polygon
• show_items - whether to display items for this polygon

• show_label and show_items are intended to be mutually exclusive per polygon, default behavior:

  • default show_label=NA and show_items=NA each row of label_df
  • function argument decides behavior for rows with NA.
  • if any show_label=TRUE for a polygon in label_df then set show_items=FALSE, or in future move the label outside the polygon.
  • if show_items=TRUE then if length(items) > max_items then set show_items=FALSE and show_label=TRUE.
  • if both show_labels=TRUE and show_items=TRUE then display both, trust the user to adjust coordinates as needed.

jmw86069/venndir documentation built on June 15, 2024, 1:52 p.m.