Workflows
End-to-end drug discovery workflows and copy-and-adapt prompt recipes, grouped by goal, each naming the tool chain it runs.
A workflow chains several platform tools into one pass from question to answer. It comes in two forms, and both are on this page:
- Guided workflows walk through the tool chain, prerequisites, step-by-step instructions, expected outputs, and best practices for one stage of discovery.
- Prompt recipes are worked examples: one prompt, the chain it triggers, and the gotchas to watch. Copy the prompt, swap in your own compound or target, and paste it into Chat.
Where Which tool? maps a single goal to a single tool, a workflow shows the whole chain end to end. You don't have to name or order the tools yourself.
Where to start
If you have a target sequence and want the classic structure-first pass, take Fold, find pockets, dock, triage. If you already have candidate compounds and need to advance them, Hit to lead is the medicinal-chemistry pipeline, and Lead optimization is the guided version. For a focused one-shot lookup, pick the recipe that names your question.
Structure and screening
Fold, find pockets, dock, triage
Recipe: from a sequence to a druggable pocket with a verified docking pose and an ADMET read — the standard structure-first entry point.
Virtual Screening
Guided: dock compound libraries against protein targets.
Sequence Analysis
Guided: protein homology, structure prediction, and annotation.
Screening Decks
Guided: build diverse screening libraries.
Hit to lead and ADMET
Hit to lead
Recipe: dock a hit, rescore the pose, screen ADMET, plan a synthesis route, and check vendor availability.
Target to lead
Recipe prompt set: from target structure to a docked, synthesizable starting point — including a generative-design variant.
Lead Optimization
Guided: hit-to-lead refinement with multi-parameter optimization.
ADMET Optimization
Guided: iterative ADMET property improvement.
Screen ADMET
Recipe: absorption, distribution, metabolism, excretion, and toxicity in one screen.
Triage a compound series
Recipe prompt set: descriptors, filters, and ADMET ranking for a series — triage, off-target, and FAERS variants.
Filter for drug-likeness
Recipe: apply medicinal-chemistry ADMET filters to a compound set.
Safety and clinical evidence
Safety Assessment
Guided: toxicity prediction and adverse event analysis.
Head-to-head safety comparison
Recipe: compare two or three drugs on off-targets, real-world adverse-event signals, and pharmacogenomics in one pass.
Compound safety profile
Recipe prompt set: off-target map, drug–drug interactions, and tox endpoint read across safety-profile, DDI, and toxicology chips.
Find clinical trials
Recipe: search trial registries for a target or indication.
Chemistry and sourcing
Search compounds
Recipe: bioactivity lookups across PubChem and ChEMBL.
Synthesis Planning
Guided: retrosynthesis routes and building block sourcing.
Compound Procurement
Guided: find and order compounds from global vendors.
Development and formulation
Polymorph Screening
Guided: crystal form prediction and stability ranking.
Formulation Development
Guided: ADMET-guided formulation decisions.
Expression Optimization
Guided: solubility prediction for construct design.
Omics and target discovery
Pathway Analysis
Guided: gene signatures, pathway enrichment, and mechanism mapping.
Discover targets from omics
Recipe: move from expression data to candidate targets.
Analyze single-cell data
Recipe: differential expression and enrichment on single-cell data.
Annotate the metabolome
Recipe: annotate metabolites and map them to pathways.
Find binding motifs
Recipe: motif discovery from ChIP-seq data.
Quality
Where the recipe prompts come from
Every <TryPrompt> block on a recipe page is copied verbatim from src/lib/orchestrator/tools/quickActionConfig.ts (the QUICK_ACTION_CHIP_POOL prompts) or src/copy/chatOnboardingRegistry.ts, so it triggers the same routing a quick-action chip would. node scripts/docs/audit-tryprompt-verbatim.mjs --strict reports any drift.
Phrasing is up to you
Every recipe prompt names tools for clarity, but you don't have to. The orchestrator infers the same chain from a plain-language goal. See How the orchestrator picks tools.
Read next
Was this page helpful?