A2A agent card imported from mrdulasolutions/MCAS.Opensource (
.well-known/agent.json). Route to it with the OpenSmartRoute A2A adapter; copyright stays with the author.
OpenMCAS Hypothesis Agent
An open, MIT-licensed hypothesis-generation agent for MCAS / MCAD (Mast Cell Activation Syndrome / Disorder). Curates a compound library across pharma, herbs, supplements, and AI-generated analogs; runs ligand-based virtual screening, covalent-warhead detection, ADMET QSAR, and multi-objective ranking across rescue / maintenance / remission categories. Every prediction is published openly so any researcher can audit, falsify, or extend it.
Skills
- Search the MCAS compound library - Search the curated MCAS/MCAD compound library by name, PubChem CID, category (rescue / maintenance / remission), target, mechanism, or evidence level. Returns matched entries with SMILES, mechanism, and evidence.
- Propose a new compound for the library - Propose a pharma drug, herb, supplement, or biologic for the curated library. Validates the PubChem CID, fetches the canonical SMILES, and opens a GitHub PR to data/compounds/seeds.json with the proposed entry.
- Report an MCAS trigger pattern - Submit an anonymous trigger pattern (food / smell / physical / stress / infection / toxin / hormonal / medication) for inclusion in the trigger -> pathway -> counter-compound map. No PHI is collected; only the pattern + onset + duration + dominant symptom domain.
- Ligand-based virtual screening against MCAS targets - Given one or more SMILES, compute max Tanimoto similarity (Morgan radius 2, 2048 bits) against curated reference ligand sets for 8 MCAS targets: MRGPRX2, KIT, KEAP1, HRH1, HRH2, CYSLTR1, BTK, GLP1R.
- Covalent warhead + KEAP1 pharmacophore scoring - Apply 13 cysteine-reactive SMARTS patterns (isothiocyanate, Michael acceptor, cyanoenone, acrylamide, vinyl sulfone, epoxide, quinone, etc.) plus a KEAP1 pharmacophore filter (MW, cLogP, heavy atoms). Returns warhead presence + KEAP1 pharmacophore pass + a [0,1] warhead score.
- ADMET QSAR prediction (hERG, AMES, BBB) - Predict cardiac liability (hERG), mutagenicity (AMES), and blood-brain barrier penetration (BBB_Martins) for a SMILES using RandomForest classifiers trained on PyTDC tasks (validation AUC 0.89-0.91).
- Generate SFN-class candidate analogs - Enumerate candidate analogs of the sulforaphane class using RDKit BRICS recombination + bioisosteric substitution + warhead grafting onto library scaffolds. Seeds: SFN, iberin, erucin, sulforaphene, allyl-ITC, benzyl-ITC, phenethyl-ITC. Returns ranked SMILES with QED + similarity + Lipinski + composite score.
- Multi-objective ranking across rescue / maintenance / remission - Compute the composite hypothesis-ranking signal: 0.30 evidence + 0.35 weighted target similarity + 0.10 QED (generated only) + 0.10 warhead score + 0.15 safety bonus + small contextual BBB term. Returns ranked CSVs per category and auto-updates the Top-10 tables in hypotheses/*.md.
- Create a standardized experiment report - Scaffold a new experiments/EXP-NNN-*.md report from the standard 11-section template, prefilled with hypothesis, method, inputs, parameters, environment, outputs, interpretation, reproduction, limitations, and next-experiment suggestions. Updates the experiments index.
- Propose or critique a hypothesis - Submit a falsifiable hypothesis (rescue / maintenance / remission / injury / trigger / cross-cutting) with falsifiable predictions, linked compounds and triggers, proposed in-silico tests, and proposed wet-lab tests. Opens a GitHub issue using the hypothesis_proposal template.
Capabilities
extensions
Endpoint: https://github.com/mrdulasolutions/MCAS.Opensource