BioMed Skill Router

Find the most suitable skill for your biomedical task by analyzing your request and matching it against available skills.

When to Use

• User describes a biomedical task but doesn't know which skill to use
• User wants to discover what skills are available for a particular domain
• User needs help choosing between multiple skills for a similar task
• User asks "What can I do with OpenBioMed?" or "Which skill should I use for X?"

How It Works

The agent analyzes the user's natural language request and determines the most suitable skill based on:

1. Understanding the user's intent and goal
2. Identifying what data/inputs the user has
3. Matching against available skill capabilities

Available Skills

Biochemistry & Drug Discovery

Skill	Description
drug-candidate-discovery	Generate diverse druggable molecules for a given target or disease using AI-powered drug discovery tools.
drug-lead-analysis	Analyze drug candidate molecules for drug-likeness (QED, Lipinski), ADMET properties, BBB penetration, and safety profiles.
target-based-lead-design	Generate diverse lead compounds for a specific protein target using structure-based drug design with MolCraft.
admet-prediction	Predict comprehensive ADMET properties (BBB penetration, side effects, Caco-2 permeability, half-life, LD50 toxicity).
retrosynthesis-planning	Expert-in-the-loop retrosynthetic planning workflow for breaking down target molecules into available starting materials.
drug-drug-interaction-analysis	Analyze potential drug-drug interactions (DDI) for up to 5 drugs using KEGG DDI database.
text-based-molecule-editing	Modify molecules based on natural language descriptions using MolT5/BioT5 models.

Protein Analysis & Engineering

Skill	Description
protein-mutation-analysis	Analyze protein mutations by retrieving protein data, explaining mutation effects with MutaPLM, and predicting structure.
mutation-design-aav	Design high-fitness and high-diversity mutants of AAV VP1 capsid protein.
mutation-design-gfp	Design high-fluorescence and high-diversity GFP mutants.
functional-protein-design	Generate functional protein sequences using CodeFun with Gene Ontology (GO) tag guidance.
similar-protein-retrieval	Retrieve proteins with similar structures or sequences from UniProt, PDB, and AFDB databases.
structure-prediction-boltz-2	Predict protein complex structures and protein-ligand complexes with binding affinity using Boltz-2.
protein-structure-design-boltzgen	All-atom protein design using BoltzGen diffusion model for binder design and peptide design.
antibody-structure-prediction-tfold	Predict antibody/nanobody structures and antigen-antibody complex structures.
antibody-design-iggm	Epitope-conditioned de novo antibody design and affinity maturation.
binding-affinity-prediction-prodigy	Predict binding affinity scores for protein complexes using Prodigy.
protein-ligand-binding-analysis-plip	Analyze protein-ligand interactions in PDB structures using PLIP (Protein-Ligand Interaction Profiler).
protein-function-prediction	Predict protein function and properties from amino acid sequence using BioT5.
protein-subcellular-localization-prediction-biot5	Predict protein subcellular localization from amino acid sequence using BioT5.

Single-Cell Omics Data Analysis

Skill	Description
single-cell-foundation-model-scrna-seq-geneformer	Geneformer workflows for tokenization, cell/gene classification, embedding extraction, and perturbation analysis.
single-cell-foundation-model-scrna-seq-langcell	LangCell for zero-shot and few-shot cell type annotation with multimodal cell-text matching.
single-cell-foundation-model-scrna-seq-scgpt	scGPT for preprocessing, binning, cell embedding extraction, fine-tuning, and reference mapping.
spatial-transcriptomics-foundation-model-stofm	SToFM for spatial transcriptomics preprocessing and cell embedding generation.
single-cell-scrna-seq-analysis-scanpy	Complete scRNA-seq analysis workflow with Scanpy including QC, normalization, clustering, and marker gene identification.
single-cell-multi-omics-analysis-scvi	Probabilistic deep learning for single-cell multi-omics analysis including scVI, scANVI, totalVI.
cellxgene-census-query	Query CZ CELLxGENE Census (61M+ cells) for single-cell expression data.
spatial-transcriptomics-spatial-data-io	Load spatial transcriptomics data from Visium, Xenium, MERFISH, Slide-seq, and other platforms.
single-cell-atac-seq-qc-processing	Trim adapters, align reads, remove duplicates, and evaluate chromatin accessibility data quality.
single-cell-atac-seq-peak-calling-annotaion	Call accessible chromatin peaks with MACS2 and identify differentially accessible regions.
single-cell-proteomics-data-processing	Load, inspect, centroid, and extract features from raw LC-MS/MS data files using pyOpenMS.
single-cell-proteomics-peptide-identification	Search MS2 spectra against protein databases with MSFragger/Comet.
single-cell-multi-omics-data-harmonization	Prepare multi-omics datasets for joint integration with normalization and batch correction.

Data Retrieval & Knowledge

Skill	Description
pubchem-query	Query PubChem database for chemical structures, similar compounds, and bioactivity data.
biomedical-literature-search	Search PubMed and bioRxiv for biomedical research papers.
uniprot-query	Query UniProt database for protein sequences, metadata, and search by criteria.
kegg-query	Query KEGG database for drug information, pathway analysis, and disease-drug-target discovery.
ppi-string-query	Query STRING database for protein-protein interactions with confidence scores.
chembl-query	Query ChEMBL database for bioactivity data on drug-like compounds.

Molecule Analysis

Skill	Description
iupac-name-identification-biot5	Identify the IUPAC name of a molecule using BioT5 question answering model.
molecule-biochemical-significance-query-biot5	Query a molecule's biochemical significance and roles in biology and chemistry.

Disease & Target Analysis

Skill	Description
disease-drug-intelligence	Comprehensive disease-to-innovative-drug analysis: targets, drugs, clinical progress, and mechanism trends.
target-drug-report	Generate comprehensive drug development progress reports for disease therapeutic targets.

Utilities

Skill	Description
biomed-skill-creator	Create new biomedical skills or improve existing ones.

Routing Process

When a user asks for help finding a skill:

1. Understand the request - Analyze what the user wants to accomplish
2. Identify available inputs - Determine what data the user has (molecule, protein, sequence, etc.)
3. Match to skills - Select the most appropriate skill(s) based on the task
4. Provide recommendation - Suggest the best skill and explain why

Example Interactions

User: "I want to design a drug for BCL2 protein"

Agent: Based on your request to design drugs for the BCL2 protein, I recommend:

target-based-lead-design - This skill generates diverse lead compounds for a specific protein target using structure-based drug design. You'll need a PDB ID for BCL2. Would you like me to help you find one?

---

User: "What can I use to analyze my scRNA-seq data?"

Agent: For scRNA-seq analysis, several skills are available depending on your goal:

• single-cell-scrna-seq-analysis-scanpy - Complete analysis workflow (QC, clustering, marker genes)
• single-cell-foundation-model-scrna-seq-langcell - Cell type annotation
• single-cell-foundation-model-scrna-seq-geneformer - Embedding extraction and perturbation analysis
• single-cell-foundation-model-scrna-seq-scgpt - Embedding extraction and reference mapping

What specific analysis do you want to perform?

---

User: "I have a molecule and want to check if it's toxic"

Agent: For toxicity analysis, use admet-prediction. This skill predicts comprehensive ADMET properties including toxicity (LD50), side effects, and BBB penetration. Please provide your molecule as a SMILES string.

See Also

• skills_overview.md - Full skill catalog with detailed descriptions
• biomed-skill-creator - Create new skills when none match your needs