What makes ADMET screening hard without a tool

Before you even reach a computational model, ADMET screening runs into a chicken-and-egg problem: predicting how a compound behaves in the body requires a trained model built on experimental data across thousands of known compounds, plus a way to turn your specific structure into the numeric descriptors that model needs. Traditional workflows solve this with specialized chemistry software, manual descriptor calculation, or outsourced lab assays — each with its own setup cost and turnaround time. A language model can discuss what ADMET properties mean in general, but it can’t run a structure through a validated prediction model or return a reproducible numeric estimate for your exact molecule.

Why this needs Chemical Molecular

Chemical Molecular’s ls_admet_predict tool calls the ADMET-AI model — a published, peer-reviewed prediction model — to estimate absorption, distribution, metabolism, excretion, and toxicity properties for each input SMILES, so the estimate comes from a named, externally validated model rather than a black-box guess. The same server also handles structure search and patent-linked structure lookup, so a molecule you’ve screened for ADMET can be checked against patent records without switching tools. It runs on the Open Platform infrastructure Patsnap has extended into life sciences alongside its longer-running patent and R&D data work.

Prepare your input

ADMET prediction starts from a molecular structure expressed as a SMILES string. If you’re starting from a compound name, drawing, or registry number, convert it to SMILES first using a standard chemistry toolkit or online converter. For this tutorial, the input is aspirin’s canonical SMILES: CC(=O)OC1=CC=CC=C1C(=O)O. Confirm the SMILES represents the exact tautomer or ionization state you want to evaluate — ADMET properties can vary with protonation state, so match your input to the physiological form relevant to your question.

1Connect Chemical Molecular

1. Get a Patsnap Open Platform API key.
2. Copy the connection URL or JSON config from the Connect panel on the Chemical Molecular page.
3. Add the server to your AI assistant or coding agent, Cursor, or another MCP-compatible client, then restart it.
4. Confirm the structure search and ADMET prediction tools are available before you ask.

2Run the example query

Predict the ADMET properties for aspirin, SMILES CC(=O)OC1=CC=CC=C1C(=O)O.

Claude recognizes the request and calls ls_admet_predict with the supplied SMILES string, returning a set of calculated molecular descriptors and ADMET endpoint estimates.

3Read the returned ADMET profile

Molecular weight and logP are physicochemical descriptors that inform solubility and permeability. AMES estimates mutagenicity risk on a 0–1 scale; aspirin’s 0.05 indicates low predicted genotoxic liability. Oral bioavailability of 0.87 suggests the compound is likely well absorbed when taken orally. BBB permeability of 0.81 indicates moderate predicted penetration across the blood-brain barrier — a consideration for CNS activity or off-target effects.

4Go further with follow-up questions

Once you have the initial profile, focus your evaluation with targeted prompts:

• Check drug-likeness — “Which properties fall outside typical drug-like ranges for an oral therapeutic?”
• Test against a threshold — “Compare this BBB permeability score to a threshold of 0.5. Does the result suggest CNS exposure?”
• Compare a structural analog — predict ADMET for a related SMILES and compare bioavailability and toxicity scores to identify favorable modifications.
• Evaluate a series of candidates — submit SMILES strings for multiple lead compounds in sequence and build a comparative table to support prioritization.

Each follow-up uses the returned values as context, so you can flag liabilities or explore structure-property relationships without leaving the conversation. Need to check the same molecule against patent records? The same server’s structure-search tools handle that too.