Interactive Planner

The Interactive Path Planner lets you retrosynthesize a target molecule one step at a time, reviewing and choosing between AI-suggested disconnections at every stage — as opposed to Tree Builder (see Tree Builder), which searches many steps automatically without you steering each one.

Access

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

Opening the Interactive Planner

1. Click Chemistry Tools in the sidebar — it opens in a new browser tab.
2. From the tools menu, open Interactive Path Planner (sometimes labeled Interactive Planning/Tree Builder — Tree Builder lives in the same module, under its own Build Tree action).

1. Enter your target molecule

Enter the target molecule as a SMILES string, or use the built-in structure editor (Ketcher) if you'd rather draw it than type SMILES by hand.

Example: CCCCC(C)(C)O (2-methyl-2-hexanol).

2. Check strategy settings

Before expanding, it's worth checking Strategy Settings — the defaults work well for most molecules, but you can tune them for harder cases:

Setting	Default	What it controls
Minimum plausibility	0.001	Filters out precursor suggestions the model scores as very unlikely. Raise it to see only the most confident suggestions.
Prediction model	template_relevance	Which retrosynthesis model proposes disconnections.
Template set	reaxys	Which reaction-template database backs the predictions.
Max. templates	1000	Upper bound on templates considered per expansion.
Max. cumulative probability	0.999	Caps how much of the model's probability mass is explored before stopping.

You generally don't need to change these for a first attempt — the defaults are a sensible starting point.

3. Expand one step

Click One Step (sometimes shown as Expand). The planner proposes the top retrosynthetic disconnections for your target and draws them as a tree/graph:

• Edge thickness reflects how strongly the model scores each precursor set — a thicker line means a more confident suggestion.
• $ markers on a chemical node mean that precursor is commercially available (buyable) in the backend database — a good sign you've reached a usable starting material.
• Reaction nodes (circles) and chemical nodes are both clickable.

4. Expand further, node by node

To keep going down a specific branch:

1. Click the chemical node you want to break down further.
2. Click Expand Node.
3. The canvas updates in place with that node's own precursor suggestions.

Repeat for as many branches as you want to explore — you're steering the search yourself, one disconnection at a time, rather than letting an automated search decide the whole route.

5. Optional: clustering

If a single expansion returns many similar-looking precursor suggestions, Strategy Settings → Clustering can group them so you see one representative per cluster instead of a wall of near-duplicates:

• Multiple clustering methods are available (e.g. by reaction class).
• With Group Cluster enabled, the canvas shows the top precursor from each of the top clusters, instead of a flat top-5 list.

This is optional — most targets don't need it, but it helps when a step has many structurally similar options.

6. Canvas controls

• Switch between tree and graph layout.
• Center the view if you've panned/zoomed away from the action.
• Screenshot/export the current canvas view.

7. Save your results

Click Save Results to keep what you've built — either download it or store it in your ASKCOS account so you can come back to it later without redoing the search.

Tips

• Start with One Step on the target itself before diving into any one branch — seeing all the top-level options first usually makes it easier to pick a sensible direction.
• A $ on a node is the fastest way to tell "this branch is close to done" from "this branch needs more expansion."
• If a branch stalls (no good precursors, low plausibility everywhere), that's useful information too — it may mean the target needs a different disconnection strategy, which is exactly the kind of judgment call this tool is built to support interactively (versus Tree Builder's fully automated search).

Troubleshooting

Symptom	Likely cause
No suggestions appear after One Step	Try lowering the minimum plausibility threshold, or double-check the SMILES is valid.
A node won't expand	Click directly on the chemical structure/node, not the surrounding canvas, then click Expand Node.
Results seem to hang	Complex targets can take a little longer than simple ones — give it a moment before retrying.

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

---

Advanced: customizing the search

Once you're comfortable with the basic one-step-at-a-time flow above, Strategy Settings has more depth worth knowing about for harder targets.

Min. Plausibility — get more suggestions, not fewer

The plausibility filter scores each proposed reaction on how chemically reasonable it looks, without factoring in reaction conditions. If a target isn't returning useful suggestions, try setting Min. Plausibility to 0 — this maximizes the number of returned suggestions, including lower-confidence ones the default threshold would normally hide. It trades precision for recall: more options to look through, but more noise to filter mentally.

Regio-selectivity checking

Turning on regio-selectivity checking adds a visual flag to reaction nodes that could plausibly give more than one product:

• Red circle — the reaction has two or more regio-selective (positional) product possibilities.
• Yellow circle — the tool can't verify regio-selectivity one way or the other for that reaction.

This is a useful early-warning signal before you commit to a branch: a red or yellow flag means "double-check this step's selectivity" rather than take the drawn product at face value.

Choosing a prediction strategy

Two different underlying strategies are available, and they can be combined:

• Template-based (template_relevance) — the default. Backed by a library of known reaction templates; you can choose which template set(s) to draw from, cap the maximum number of templates considered, and cap the maximum cumulative probability explored before the search stops. Multiple template sets can be selected together.
• Template-free (augmented_transformer) — predicts disconnections directly, without a template library, choosing a training dataset (e.g. USPTO_FULL or a larger Pistachio-derived set) instead of a template set.

Combining both strategies at once broadens the suggestions you see beyond what either model would propose alone — useful when a target's first-pass results all look similar or unconvincing.

Per-node actions worth knowing

Right-clicking (or selecting) a chemical node opens more than just Expand Node:

• View Notes — attach your own annotation to a molecule, useful for tracking your reasoning as a route gets long.
• Add Precursor — manually add a precursor molecule yourself if the model's suggestions don't include one you know works. Useful for combining your own chemical knowledge with the model's suggestions rather than being limited to what it proposes.
• View Recommended Templates — see the specific templates behind a suggestion, including their scores, ranks, and supporting literature reactions — useful for sanity-checking why the model proposed a disconnection.
• Ban — exclude a specific chemical from consideration, if you know it's not a viable precursor (too expensive, wrong stereochemistry, etc.) and don't want the search to keep suggesting it.

Once a node is expanded, you can also sort its precursor sets by different criteria, and filter to precursors involving a specific atom you click on — handy for focusing on one part of a large molecule.

Reading reaction node details

What a reaction node shows depends on which strategy produced it:

• template_relevance nodes show score, template rank, template score, plausibility, and example count, plus the supporting templates behind the suggestion.
• augmented_transformer nodes show score and plausibility only (no template detail, since this strategy doesn't use templates).

An Evaluate Reaction button on any reaction node jumps straight into Forward Synthesis with that reaction's reactants pre-filled — a fast way to sanity-check a proposed disconnection by predicting its forward product.