Skip to content

Tree Builder

Tree Builder runs a fully automated, multi-step retrosynthesis search — using Monte Carlo Tree Search (MCTS) — to find complete synthetic routes from your target molecule back to purchasable starting materials, without you steering each individual step. If you'd rather review and choose each disconnection yourself, see Interactive Planner instead — both tools live in the same module.

Access

Chemistry Tools requires an active subscription to any subject — see Chemistry Tools Overview for the full access model.

Jobs run in the background

Tree Builder submits an asynchronous job — you don't watch it run live. Results are saved to your account, so you must be signed in for a search to work.

Opening Tree Builder

  1. Click Chemistry Tools in the sidebar — it opens in a new browser tab.
  2. Open the Interactive Path Planner / Tree Builder module, then choose the Build Tree action (as opposed to One Step, which is Interactive Planner's manual mode).

1. Enter your target molecule

Enter the target as a SMILES string, or draw it with the built-in Ketcher structure editor.

2. Check strategy settings

The same Strategy Settings as Interactive Planner apply here (see Interactive Planner for the full table) — minimum plausibility, prediction model, template set, max templates, and max cumulative probability. Defaults work well for a first attempt.

Under MCTS / Tree Builder Settings, the defaults are usually fine as-is. The one setting worth checking before a long search: the expansion time limit — a lower value (e.g. capping it around 30 seconds) keeps the search fast for a first pass; you can always rerun with a longer limit if the first pass doesn't find a satisfying route.

4. Build the tree

Click Build Tree. A popup confirms the job was submitted to the backend — the search itself runs asynchronously, so you don't need to keep the tab open and watching.

5. Check on progress

Open My Results in the sidebar to see your submitted jobs:

  • Started — the search is still running.
  • Completed — a notification appears, and the job's status updates. From here you can:
    • View in IPP — opens the found route(s) in the Interactive Path Planner canvas, so you can inspect, tweak, or expand further.
    • View Trees — browse the discovered synthetic routes directly.

Reading the results

Completed routes use color coding to show status at a glance:

  • Green boxes — purchasable (buyable) starting materials. These are the endpoints of a finished route.
  • Red boxes — compounds that still need further expansion (no buyable precursor found down that branch yet).

A route made entirely of green boxes at its leaves is a complete, buyable synthesis path back from your target.

Tips

  • Tree Builder searches take longer than One Step predictions — a full multi-iteration search naturally takes more time than a single retrosynthesis call. This is expected; let it run rather than assuming it's stuck.
  • For complex or unusual targets, the search may not find a single complete path to buyable building blocks. This is a known limitation of the search algorithm for hard targets, not a sign something is broken — try adjusting Strategy Settings (e.g. raising max templates) or switch to Interactive Planner to explore manually where the automated search struggled.
  • Because jobs run in the background, you can start a search, do something else, and come back to My Results later.

Troubleshooting

Symptom Likely cause
Job stays "started" for a long time Normal for complex targets — multi-step searches can take a while. Check back later rather than resubmitting.
No complete route found Some targets don't resolve to buyable materials within the search budget — try Interactive Planner for manual exploration, or loosen Strategy Settings.
"Build Tree" seems to do nothing Make sure you're signed in — results are tied to your account, and an expired session can silently block job submission.

Still stuck? See Chemistry Tools Overview for general troubleshooting, or email support@examcatalyst.ai.

If a default search comes back with no routes (or too few), the fix is almost always in Strategy Settings, not a sign the target is unsolvable.

The General tab controls the same one-step retrosynthesis model settings used by Interactive Planner — Min. Plausibility, template/strategy choice, and so on. Everything in that page's Advanced section applies here too, since Tree Builder repeatedly calls the same one-step model under the hood at every node of its search.

The MCTS / Tree Builder Settings tab has three parameters specific to the automated search itself:

Parameter What it controls
Expansion time How long the search keeps looking for routes. The single biggest lever — a search that returns nothing at a short default time limit often returns many routes once given more time.
Max expansion depth How many steps deep the search is allowed to go down one branch before it explores alternatives. Raise it for targets that plausibly need a longer route.
Max branching How many precursor sets the search considers at each node. Raise it to explore more alternatives per step, at the cost of a larger search space.

Why this matters, concretely: a target that returns zero synthesis trees at default settings can return dozens once Expansion time is raised to a minute — and even more once Max expansion depth and Max branching are also raised. If your first Tree Builder run comes back empty, this is the first thing to try before concluding the target has no viable route: increase expansion time first (it's the setting most likely to matter), then depth and branching if you still get nothing useful.

There's a real tradeoff here — higher values mean a longer wait for results, and a larger space of routes to review afterward. Start with a modest increase (e.g. doubling expansion time) rather than maxing out all three at once.

Advanced: analyzing your results

Once a search completes, open View Trees to enter Tree Explorer mode — this is where you go beyond just reading routes one at a time.

  • Use the arrow controls above the display to step through the returned routes. By default, routes are sorted by number of reaction steps (fewest first).
  • The Sort trees dropdown lets you choose your own primary (and secondary, tertiary) sort criteria, with a toggle for ascending/descending order.

Analyze trees

The Analyze trees panel runs additional computation on your results in the background — each button populates its results once it finishes and the view refreshes:

  • Score and Cluster Pathways — evaluates and groups similar routes together. Toggle View by cluster afterward to browse cluster-by-cluster instead of route-by-route — much easier to make sense of a large result set than scrolling through every route individually. Separate arrow controls let you move between clusters and between routes within a cluster.
  • Reaction Classification — labels each reaction step in every route with its reaction class, added directly to that reaction node's details. Useful for spotting, at a glance, which steps in a route are familiar named reactions versus less common transformations.
  • Process Mass Intensity (PMI) — estimates the mass-efficiency of a route (how much total input mass is needed per unit of product). Run it for a single tree or across all returned trees at once; the average PMI shows in the top-left corner of the view. Lower PMI means a more mass-efficient route.
  • Count Analogs — estimates how many structural analogs of your target molecule are also synthetically accessible via a similar route, cross-referenced against ASKCOS's buyable-compounds database. Shown in the top-left corner.

Running PMI and/or Count Analogs unlocks them as additional options in the Sort trees dropdown — so once you've computed them, you can re-sort your whole result set by mass-efficiency or analog count, not just by step count.

Suggested workflow for a large result set

  1. Run Score and Cluster Pathways first if the search returned many routes — it's the fastest way to stop scrolling through near-duplicate routes one by one.
  2. Run PMI if efficiency matters for your comparison (e.g. choosing between two otherwise-similar routes).
  3. Use Reaction Classification to quickly spot which routes use well-precedented reaction types versus more exotic ones — often a useful tiebreaker.