Manhattan-Like chart generation

[22]

Hi all,

I need a hand in producing a bar chart that is Manhattan-like. X-axis is one of the variables in my dataset and Y-axis should be generated automatically with a suitable low/high limits where these limits would be again taken from the values of the variables represented by the bar chart.

I tried to attach an Excel sheet that shows exactly what I mean (3 series of data [3 variables PCT_222, PCT_233, PCT_244) all together in the same plot and their X-axis is a variable called (Class) while their Y-axis is generated automatically using their values all together), but I was unable to due to the fact that the third shows .xlsx as invalid extension. Therefore, I attached a PDF that is generated by the Excel sheet.

I'm not sure if this is direct in SAS or not and I hope so.I searched before writing here but could not find any useful resource which shows somthing similar to my requirement and a straight answer to it. Therefore, I hope the Graph/ODS experts here can assit in resolving this problem which might turned to be a trivial one.

Regards

[22]

Just to add, I'm using SAS 9.3

[BrunoMueller · Accepted solution]

You should have a look at ODS Graphics. @Jay54 maintains a blog with lots of great examples.

To get you started find below some sample code. I have not tested it with SAS 9.3, you have a go.

data have;
  infile cards dlm=",";
  input
    Class
    Overall pct_overall : percent10.
    V_222 PCT_222 : percent10.
    V_233 PCT_233 : percent10.
    V_244  PCT_244: percent10.
  ;
  format pct_: percent10.2;
  cards;
4,3,1%,0,0%,0,0%,3,2%
5,22,15%,60,45%,31,15%,40,30%
7,50,35%,58,38%,44,33%,33,40%
8,130,27%,77,32%,50,40%,20,15%
9,70,18%,18,13%,15,11%,18,9%
10,33,8%,13,8%,17,8%,9,7%
13,9,4%,6,4%,25,20%,7,4%
14,11,2%,2,1%,7,3%,1,1%
16,18,6%,0,0%,9,8%,4,5%
17,12,3%,0,0%,4,2%,5,5%
19,8,2%,0,0%,6,3%,1,1%
20,6,1%,0,0%,3,1%,2,3%
;

proc sgplot data=have;
  vbar class / response=pct_overall barwidth=0.2 discreteoffset=-0.3
    legendlabel="Overall" fillattrs=(color=cxffff00)
  ;
  vbar class / response=pct_222 barwidth=0.2 discreteoffset=-0.1
      legendlabel="V_222 "
  ;
  vbar class / response=pct_233 barwidth=0.2 discreteoffset=0.1
      legendlabel="V_233"
  ;
  vbar class / response=pct_244 barwidth=0.2 discreteoffset=0.3
      legendlabel="V_244"
  ;
  keylegend / noborder across=1 location=inside  position=right;
  yaxis label="Account(%)";
run;



proc transpose 
  data=have
  out=have_trsp
;
  by Class;
  var pct_:;
run;

proc sgplot data=have_trsp;
  vbar class /
    response=col1 group=_name_ groupdisplay=cluster GROUPORDER=DATA
  ;
  yaxis label="Account(%)";
  keylegend / 
    across=1 noborder location=inside  position=right title=""
  ;
run;

Bruno

[22]

Hi Bruno,

Thanks for your time and I really appreciate your professional answer as a complete solution including recreating the data and perform all necessary steps to reach to proc sgplot. In addition, doing the sgplot using two approaches (though I liked the transposed data as it shortened the code even with the extra step of proc transpose).

Thanks a lot that is exactly what I want. 

Last small question, Can I control the vbar colors for each group?

Regards

[BrunoMueller]

Have a look here, with SAS 9.4 it is very easy, with SAS 9.3, there is some more effort

Bruno

[MikeZdeb]

Hi.  Not sure what you mean by a Manhattan plot since your exaple does not look like what I understand such a plot to be.

If you want a standard Manhattan plot, take a look at ...

Creating Manhattan Plots with SAS/Graph

http://www.sascommunity.org/wiki/Creating_Manhattan_Plots_with_SAS/Graph

First paragraph in this link ...

A Manhattan plot is a type of bar chart that is sometimes used in the visualization of genome-wide association data. Some examples can be found by using Google to search for "Manhattan plots" then clicking on "Images" on the upper left-hand side of the search results. You can see Manhattan plots used in publications by clicking on either of the following links: example 1 or example 2.

has couple links to Manhattan plots in publications.

[MikeZdeb]

Hi.  Here's an example using GPLOT (admittedly more stuff than SGPLOT).  It produced the attached chart (default color scheme used) ...

* your data;
data x;
input class overall pct_over v_222 pct_222 v_233 pct_233 v_244 pct_244;
datalines;
4 3 1 0 0 0 0 3 2
5 22 15 60 45 31 15 40 30
7 50 35 58 38 44 33 33 40
8 130 27 77 32 50 40 20 15
9 70 18 18 13 15 11 18 9
10 33 8 13 8 17 8 9 7
13 9 4 6 4 25 20 7 4
14 11 2 2 1 7 3 1 1
16 18 6 0 0 9 8 4 5
17 12 3 0 0 4 2 5 5
19 8 2 0 0 6 3 1 1
20 6 1 0 0 3 1 2 3
;

* rearrange your data for use in PROC GCHART;
* three variables in new data set;
* CLASS;
* _NAME_ (the name of the PCT variable from data set x);
* COL1 (values of percentages in data set x);
proc transpose data=x out=y;
by class;
var pct: ;
run;
 
* rescale the percentages from 0-100 to 0-1 (so you can use the PERCENT format in GCHART);
data y;
set y;
col1 = col1/100;
run;
 
* specify a destination for the PNG file produced by GCHART;
filename gout 'z:\chart.png';
 
* specify details for the chart;
goptions reset=all gunit=pct ftext='calibri' htext=2 gsfname=gout

         dev=png xpixels=1500 ypixels=1000;

* override the default axis look in GCHART;
axis1 label=none value=none;
axis2 label=none offset=(2,2)pct;
axis3 label=(h=3 a=90 'ACCOUNT(%)') order=0 to .50 by .05 minor=none;

* move the legend inside the chart frame and tweak defaults;
legend1 position=(top inside right) across=1 label=none shape=bar(3,4)
         value=('V_222' 'V_233' 'V_244' 'OVERALL') mode=protect offset=(-5,-5)pct;

* adds whitespace at top;
title1 ls=2;
* adds a centered label for the horizontal access;
footnote1 h=3 j=c 'CLASS';

* draw the chart (produces a PNG file);

proc gchart data=y;
vbar _name_ / subgroup=_name_ sumvar=col1 group=class
maxis=axis1 gaxis=axis2 raxis=axis3
space=0 legend=legend1;
format col1 percent.;
run;
quit;

[22]

Hi MikeZdeb,

I'd like to thank you too for the professional answer you provided. This has also an added value as you've used different proc (proc gchart) which did a similar result but with a slightly different way.

Thanks a lot MikeZdeb. I'd also consider this as a full answer to my query.