Reaction tree plain text visualization

[tim-hsu]

I'm not sure if this should be a PR since it's not an essential feature, so I'm documenting the feature here as an issue.

Example visualizations

A reaction tree is the tree representation of a chemical synthesis route, with the root node being the target molecule, which is always printed on the first line.

Tamoxifen

CC/C(=C(\c1ccccc1)c1ccc(OCCN(C)C)cc1)c1ccccc1
├── CC/C(=C\c1ccccc1)c1ccccc1
│   └── CCC(O)(Cc1ccccc1)c1ccccc1
│       ├── CCC(=O)Cc1ccccc1
│       └── Brc1ccccc1
└── CN(C)CCOc1ccc(I)cc1
    ├── CN(C)CCCl
    └── Oc1ccc(I)cc1

Ritonavir

CC(C)c1nc(CN(C)C(=O)N[C@H](C(=O)N[C@@H](Cc2ccccc2)C[C@H](O)[C@H](Cc2ccccc2)NC(=O)OCc2cncs2)C(C)C)cs1
├── CC(C)c1nc(CN(C)C(=O)N[C@H](C(=O)Cl)C(C)C)cs1
│   ├── O=C(Cl)C(=O)Cl
│   └── CC(C)c1nc(CN(C)C(=O)N[C@H](C(=O)O)C(C)C)cs1
│       └── COC(=O)[C@@H](NC(=O)N(C)Cc1csc(C(C)C)n1)C(C)C
│           ├── CNCc1csc(C(C)C)n1
│           └── COC(=O)[C@@H](NC(=O)Oc1ccc([N+](=O)[O-])cc1)C(C)C
│               ├── COC(=O)[C@@H](N)C(C)C
│               └── O=C(Cl)Oc1ccc([N+](=O)[O-])cc1
└── N[C@@H](Cc1ccccc1)C[C@H](O)[C@H](Cc1ccccc1)NC(=O)OCc1cncs1
    └── [N-]=[N+]=N[C@@H](Cc1ccccc1)C[C@H](O)[C@H](Cc1ccccc1)NC(=O)OCc1cncs1
        └── CC(C)(C)[Si](C)(C)O[C@@H](C[C@H](Cc1ccccc1)N=[N+]=[N-])[C@H](Cc1ccccc1)NC(=O)OCc1cncs1
            ├── O=C(OCc1cncs1)Oc1ccc([N+](=O)[O-])cc1
            │   ├── OCc1cncs1
            │   └── O=C(Cl)Oc1ccc([N+](=O)[O-])cc1
            └── CC(C)(C)[Si](C)(C)O[C@@H](C[C@H](Cc1ccccc1)N=[N+]=[N-])[C@@H](N)Cc1ccccc1

In-house implementation

The in-house code is fairly simple, as shown below.

How to reproduce

Add this function to the ReactionPath class:

def show_tree(self, stdout: bool = True) -> str | None:
    # Outer, nonlocal string to be modified during depth-first search
    # In this string, the first line is always the root node
    string = self.root.smiles + '\n'

    # Depth-first search
    def dfs(curr: Node, prefix: str):
        nonlocal string
        for i in range(len(curr.children)):
            node_id = curr.children[i]
            if i == len(curr.children) - 1:
                string += prefix + '└── ' + self.nodes[node_id].smiles + '\n'
                dfs(self.nodes[node_id], prefix + '    ')
            else:
                string += prefix + '├── ' + self.nodes[node_id].smiles + '\n'
                dfs(self.nodes[node_id], prefix + '│   ')
    
    dfs(self.root, '')
    
    # Whether to print to console or return the string object
    if stdout:
        print(string)
    else:
        return string

Then, run the following:

from charge.servers.AiZynthTools import find_synthesis_routes

routes = find_synthesis_routes('CN(C)CCOc1ccc(cc1)/C(c2ccccc2)=C(/CC)c3ccccc3')  # tamoxifen
path = ReactionPath(routes[0])
path.show_tree()

treelib implementation

If we want more features for reaction tree analysis and visualization, I recommend using the treelib package, which is very light-weight and only requires the six dependency.

How to reproduce

from treelib import Tree
from treelib.node import Node

def route2tree(route: dict) -> Tree:
    tree = Tree()
    def dfs(curr: dict, parent: Node | None):
        if curr['type'] == 'reaction':
            for child in curr['children']:
                dfs(child, parent)
        else:
            # You can add custom data (such as `in_stock`) with the `data` keyword
            node = tree.create_node(tag=curr['smiles'], parent=parent, data={'in_stock': curr['in_stock']})
            if 'children' in curr:
                for child in curr['children']:
                    dfs(child, node)
    dfs(route, None)
    return tree

routes = find_synthesis_routes('CN(C)CCOc1ccc(cc1)/C(c2ccccc2)=C(/CC)c3ccccc3')  # tamoxifen
tree = route2tree(routes[0])
tree.show()