When trees are redrawn - the tool is not selecting the current node.

The parser text reading is not intelligent enough to differentiate the edge label from node labels and also not able to identify key and values.
Say for e.g
(expr S
(conj S [ref=4]
(subj NP [ref=1]
(head PersPron [ref=1]
(lex நான் [stem=I,case=1,person=1,number=sg])))
(head V [ref=3]
(lex தூங்குகிறேன் [stem=sleep,tense=present,person=1,number=sg,mode=active])))
...
in this lex is the edge label and நான் or தூங்குகிறேன் are node labels

Say for e.g.
(expr S
(conj S [ref=4]
(subj NP [ref=1]
(head PersPron [ref=1]
(lex நான் [stem=I,case=1,person=1,number=sg])))
(head V [ref=3]
(lex தூங்குகிறேன் [stem=sleep,tense=present,person=1,number=sg,mode=active])))

The tool produce graph visual output with node label as 'நான் [stem=I,case=1,person=1,number=sg]'. When only 'நான்' is displayed, this reduces graph size visual output, which will improve clarity in dealing with a large graph.

The idea would be to show a list with boolean flags. The list displays all the features for one graph and checking on and off with boolean flags will display or hide that particular features in the graph.

The format of the JSON needs to be changed to accommodate edge labels and features.
Right now all these information are put in as if they are part of the text node itself.