tooluniverse-drug-synergy
This tool quantifies drug-drug interaction synergy by comparing observed combination effects against five reference models (Bliss independence, HSA, Loewe additivity, ZIP, and Chou-Talalay Combination Index). Use it when you have measured single-drug and combination inhibition or viability data and need to determine whether drugs act synergistically, additively, or antagonistically, along with guidance on which model matches your experimental design.
git clone --depth 1 https://github.com/mims-harvard/ToolUniverse /tmp/tooluniverse-drug-synergy && cp -r /tmp/tooluniverse-drug-synergy/plugin/skills/tooluniverse-drug-synergy ~/.claude/skills/tooluniverse-drug-synergySKILL.md
# Drug-Combination Synergy Analysis
Decide whether a two-drug combination does **more than expected** (synergy), exactly as expected (additivity), or **less** (antagonism) — and pick the right reference model for the data you have.
"Synergy" only means something relative to a **null model of additivity**, and the models define additivity differently — so the first decision is *which model*, driven by *what data you measured*.
## Step 0 — Pick the model by the data you have
| You measured… | Use model | Tool | Input |
|---|---|---|---|
| Single effects of A, B, and A+B at **one** dose pair | **Bliss** | `DrugSynergy_calculate_bliss` | `effect_a`, `effect_b`, `effect_combination` (each a fraction 0–1) |
| Effects of A, B, A+B across **several** dose points | **HSA** | `DrugSynergy_calculate_hsa` | `effects_a`, `effects_b`, `effects_combo` (arrays) |
| Single-agent **dose-response curves** + one combination point | **Loewe** | `DrugSynergy_calculate_loewe` | `doses_a_single`/`effects_a_single`, `doses_b_single`/`effects_b_single`, `dose_a_combo`, `dose_b_combo`, `effect_combo` |
| Single-agent dose-response + combo point, want Chou-Talalay CI | **Combination Index** | `DrugSynergy_calculate_ci` | same as Loewe + `assumption` |
| A full **dose × dose viability matrix** | **ZIP** | `DrugSynergy_calculate_zip` | `doses_a`, `doses_b`, `viability_matrix` (% , 0–100) |
> **Effects must be on a consistent inhibition scale.** Bliss/HSA/Loewe expect *fractional inhibition* `0–1` (0 = no effect, 1 = complete kill). If your data is % viability, convert: `inhibition = 1 − viability/100`. ZIP takes the viability matrix in % directly. Mixing scales is the most common error.
## Step 1 — What each model's "additivity" means
| Model | Null (additive) expectation | Best when |
|---|---|---|
| **Bliss independence** | drugs act independently: `E_exp = E_a + E_b − E_a·E_b` | different mechanisms; quick single-point screen |
| **HSA (highest single agent)** | combo should beat the *better* single agent: `E_exp = max(E_a, E_b)` | conservative "does it beat monotherapy?" question |
| **Loewe additivity** | a drug combined with itself = additive (dose equivalence) | same/similar mechanism; needs dose-response |
| **ZIP** | combines Bliss + Loewe; potency shift of one drug's curve by the other | dose-matrix screens (the SynergyFinder default) |
| **Chou-Talalay CI** | CI<1 synergy, =1 additive, >1 antagonism (median-effect) | classic isobologram-style analysis with dose-response |
There is no single "correct" model — **state which one you used.** Bliss and Loewe genuinely disagree for some combinations (that's expected, not an error); reporting two models (e.g. Bliss + HSA, or Loewe + ZIP) is good practice.
## Step 2 — Run it
```bash
# Bliss (single dose pair, fractional inhibition)
tu run DrugSynergy_calculate_bliss '{"operation":"calculate_bliss",
"effect_a":0.4,"effect_b":0.3,"effect_combination":0.7}'
# -> expected 0.58, bliss_synergy_score 0.12, "Strong synergy"
```
`scripts/synergy_reference.py` computes the Bliss, HSA, and Loewe-style expected combination effects side-by-side from one dose pair, so you can see at a glance whether the models agree before running the full tools.
## Step 3 — Interpret the score
For Bliss/HSA/Loewe/ZIP, the synergy score is `(observed − expected)` (often ×100):
| Score (fractional, ×100 scale) | Call |
|---|---|
| > +10 | synergy |
| −10 to +10 | additive (no meaningful interaction) |
| < −10 | antagonism |
For **Combination Index (Chou-Talalay)**: **CI < 1 = synergy**, CI = 1 additive, CI > 1 antagonism (note the opposite direction — lower is more synergistic).
- A positive Bliss/HSA score means the combination exceeds the additive expectation at that point.
- Synergy is often **dose-dependent** — a combination can be synergistic at one ratio and antagonistic at another; for a matrix, report the synergistic *region*, not one number.
## Step 4 — Gotchas (state these)
- **Scale mismatch** (% viability vs fractional inhibition) — convert first (Step 0).
- **Effects near 0 or 1 (ceiling).** If both single agents already kill ~everything, the combo can't show synergy (no headroom) — Bliss/HSA saturate; interpret with care.
- **ZIP/Loewe/CI need real dose-response** with ≥3 non-zero, measurable-effect dose points per drug, or the Hill fit fails (the tools say so).
- **Model disagreement is normal** — don't shop for the model that gives "synergy"; pre-specify the model and report it.
- **A synergy score is not efficacy** — a strongly synergistic combination can still be weak overall; report the absolute combination effect too.
## Honest limitations
- These are *reference-model* synergy scores, not statistical tests — for confidence, replicate and report variability across the dose matrix.
- Synergy in vitro does not guarantee clinical benefit (PK/PD, toxicity, scheduling all matter).
## Related skills
- `tooluniverse-dose-response` — fit the single-agent IC50/EC50 curves that Loewe/CI/ZIP need.
- `tooluniverse-cell-line-profiling` — look up *pre-computed* combination synergy (SYNERGxDB).
- `tooluniverse-drug-repurposing` / `tooluniverse-network-pharmacology` — rationale for combinations.Install and configure ToolUniverse for any use case — MCP server (chat-based), CLI (command line with 9 subcommands), or Python SDK (Coding API with 3 calling patterns). Covers uv/uvx setup, MCP configuration for 12+ AI clients (Cursor, Claude Desktop, Windsurf, VS Code, Codex, Gemini CLI, Trae, Cline, etc.), full CLI reference (tu list/grep/find/info/run/test/status/build/serve), Coding API quickstart, agentic tools, code executor, API key walkthrough, skill installation, and upgrading. Use when user asks how to set up ToolUniverse, which access mode to use (MCP vs CLI vs SDK), configuring MCP servers, using the CLI, troubleshooting installation, upgrading, or mentions installing ToolUniverse or setting up scientific tools. Also triggers for "how do I use ToolUniverse", "what's the best way to access tools", "command line", "tu command", "coding API", "tu build".
Systematic ACMG/AMP germline variant classification with all 28 criteria (PVS1, PS1-4, PM1-6, PP1-5, BA1, BS1-4, BP1-7) for clinical significance. Produces 5-tier verdict (Pathogenic / Likely Pathogenic / VUS / Likely Benign / Benign) with cited evidence per criterion. Use for variant interpretation, VUS resolution, and pathogenicity assessment. Combines ClinVar, gnomAD, computational predictors, and gene-mechanism context.
Comprehensive ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) profiling for drug candidates. Integrates ADMET-AI predictions, SwissADME drug-likeness, PubChemTox experimental toxicity, ChEMBL clinical data, Lipinski rule-of-five, and CYP interaction data. Use for drug-likeness assessment, BBB penetration, bioavailability, hepatotoxicity prediction, ADME/PK profiling, or screening compound libraries before lab testing.
Detect and analyze adverse drug event signals using FDA FAERS reports, drug labels, and disproportionality statistics (PRR, ROR, IC). Generates quantitative safety signal scores (0-100) with evidence grading. Use for post-market surveillance, pharmacovigilance, drug safety assessment, regulatory submissions, and detecting rare AE signals not visible in clinical trials.
Map environmental and industrial chemicals to adverse outcome pathways (AOPs) — molecular initiating event to organ-level toxicity. Uses AOPWiki, GHS classification, IARC carcinogen status, and LD50 data. Use for environmental/industrial chemical risk assessment, regulatory-grade hazard characterization, and AOP stressor mapping. Distinct from drug-safety analysis (use tooluniverse-pharmacovigilance for drugs).
Aging biology, cellular senescence, and longevity research. Covers senescence markers (p16/CDKN2A, SASP, SA-beta-gal), aging hallmarks, senolytic drug discovery (dasatinib+quercetin, fisetin, navitoclax), epigenetic clocks, telomere biology, and longevity GWAS. Use for senescence-pathway analysis, age-related disease genetics, senolytic-target discovery, and centenarian-genetics queries. Distinguishes correlative vs causal evidence (knockout, intervention).
Therapeutic antibody engineering and optimization, lead-to-clinical-candidate. Covers sequence humanization (germline alignment, framework retention), affinity maturation, developability (aggregation, stability, PTMs), structure modeling (AlphaFold/PDB CDR analysis), immunogenicity prediction, and manufacturing feasibility. Use for biologic-drug optimization, mAb design review, biosimilar engineering, and clinical-precedent comparison.
Discover novel small-molecule binders for protein targets using structure-based and ligand-based screening. Covers druggability assessment, known-ligand mining (ChEMBL, BindingDB), similarity expansion, ADMET filtering, and synthesis feasibility. Use for hit identification, virtual screening, target-to-compounds workflows, and lead-finding before commit-to-medchem.