Feature domains and capabilities
The agent workbench handles six categories of tasks—protein query, protein engineering, protein discovery, protein generation, molecular docking, and protein properties—by combining public databases, models, and computational software to perform retrieval, analysis, or candidate design. This page helps you decide which direction a task belongs to and what materials or analyses you can request in your question.
Scope of use
The names listed in the tables describe the data sources and computational capabilities the platform may draw on; they do not mean every task calls all of them. The capabilities actually available, input requirements, run time, and results are subject to the prompts shown when you submit. Computational predictions, database records, and literature summaries are not experimental conclusions. The available feature domains and specific tools vary with the selected agent mode, workspace scope, and subscription permissions; if something is unavailable, switch modes or check your permissions.
Overview
| Feature domain | Core retrieval areas | Databases involved | Models involved | Software involved |
|---|---|---|---|---|
| Protein query | Protein sequences/annotations, structures, compounds, kinetics, literature, regulatory data | UniProt, NCBI, PDB/PDBe, ChEMBL, PubChem, SABIO-RK, PubMed, OpenFDA | — | — |
| Protein engineering | Sequence redesign, mutation optimization, humanization, energy optimization | IMGT human germline reference library | ProteinMPNN, VenusPrime/VenusRem (in-house), naturalness scoring models | Rosetta, GROMACS, ESMFold/Chai1 (refolding validation) |
| Protein discovery | Homology search, sequence alignment, candidate mining | NCBI sequence databases, RCSB PDB, UniProt | VenusMine (in-house) | NCBI BLAST, EMBOSS, SeqKit, Clustal Omega/MUSCLE |
| Protein generation | Backbone generation, sequence design, antibody/nanobody de novo design | — | RFDiffusion, RFantibody, mBER VHH, ProteinMPNN, VenusG/VenusX (in-house) | ESMFold/AlphaFold 2/Chai1/Protenix (folding validation) |
| Molecular docking | Receptor-ligand/protein-protein docking, pocket prediction, interaction analysis | PubChem, ChEMBL (ligand sources) | — | AutoDock Vina, AutoDock4, HDOCK, ZDOCK, ADCP, ALDE, PLIP, MGLTools, OpenBabel, PyMOL |
| Protein properties | Physicochemical properties, secondary structure, functional annotation, interaction networks, pKa | STRING, GO, KEGG, UniProt | PSIPRED, VenusG/VenusX (in-house) | DSSP, PROPKA, protein property calculators, GROMACS |
Protein query
Suited to looking up basic facts about known proteins or compounds and consolidating information from multiple sources into verifiable results.
| Information you can obtain | Representative sources |
|---|---|
| Sequences, functional annotations, standard names, and ID mapping | UniProt, NCBI |
| 3D structures and structure entries | PDB, RCSB PDB, PDBe |
| Compounds, SMILES, and activity data | ChEMBL, PubChem |
| Enzyme kinetics parameters and experimental conditions | SABIO-RK |
| Papers, reviews, and biomedical literature | PubMed |
| Adverse drug events and regulatory information | OpenFDA |
When asking, provide the species, UniProt ID, PDB ID, compound name, or SMILES where possible, and ask for direct experiments, database annotations, and inferences to be distinguished.
Protein engineering
Suited to proposing mutation and sequence optimization candidates toward activity, stability, solubility, affinity, or humanization goals.
| Analysis direction | Representative capabilities |
|---|---|
| Backbone-guided sequence design, loop repair, and mutation optimization | ProteinMPNN |
| Platform engineering and optimization pipelines | VenusPrime, VenusRem |
| Antibody humanization with germline references | IMGT reference system, humanization analysis |
| Naturalness screening of candidates | Naturalness scoring models |
| Energy evaluation, conformational sampling, and side-chain optimization | Rosetta |
| Energy minimization and molecular dynamics stability assessment | GROMACS |
| Refolding validation of engineered structures | ESMFold, AlphaFold 2, Chai1, Protenix |
Batch and chaining: ProteinMPNN accepts multiple PDB files or an archive in one submission and returns merged sequence results; AlphaFold2 supports MSA-free single-chain batch folding, and the two can be chained. When results exceed 200, the page previews 200 representative structures and provides the full results as an archive download.
Provide the original sequence or structure, known functional/binding sites, residues that must not change, target properties, and experimental constraints. Candidate mutations should be further validated by expression, activity, and stability experiments.
Protein discovery
Suited to finding candidate proteins from public sequence and structure resources, then filtering and ranking them by homology, conservation, or user-defined criteria.
| Analysis direction | Representative resources or tools |
|---|---|
| Homology search and candidate confirmation | NCBI BLAST, NCBI sequence databases, UniProt |
| Structure-sequence alignment and template discovery | RCSB PDB sequence alignment |
| Candidate mining | VenusMine |
| Multiple sequence alignment and conservation analysis | EMBOSS, Clustal Omega, MUSCLE |
| Sequence quality control, filtering, deduplication, and extraction | SeqKit |
State the target function, species or source scope, thresholds (such as E-value or coverage), number of candidates, and the evidence criteria used for ranking.
Protein generation
Suited to generating protein backbones or sequences from scratch and designing candidates such as antibodies and nanobodies. Generated results usually require folding prediction, structural checks, and downstream experiments for screening.
| Analysis direction | Representative models or capabilities |
|---|---|
| Diffusion-model backbone generation and sampling | RFDiffusion |
| Antibody design | RFantibody |
| Nanobody de novo design | mBER VHH |
| Constraint-guided sequence generation | ProteinMPNN |
| Platform protein generation and design | VenusG, VenusX |
| Structure prediction with confidence assessment | ESMFold, AlphaFold 2, Chai1, Protenix |
Clarify the target epitope or structural constraints, desired function, allowed sequence/structure range, number of candidates, and the validation and screening criteria you want applied.
Molecular docking
Suited to analyzing how receptors may bind small molecules, peptides, or proteins, including pocket prediction, conformation generation, and interaction checks.
| Analysis direction | Representative resources or tools |
|---|---|
| Ligand retrieval and acquisition | PubChem, ChEMBL |
| Small molecule-protein docking and scoring | AutoDock Vina, AutoDock4 |
| Protein-protein docking | HDOCK, ZDOCK |
| Additional docking engines | ADCP, ALDE |
| Pocket prediction and binding box definition | ProteinPlus, PyMOL |
| Receptor/ligand preparation and format handling | MGLTools, OpenBabel |
| Hydrogen bond, salt bridge, and π–π interaction analysis | PLIP |
| Docking result visualization | PyMOL |
Interaction analysis supports uploading a complex structure or providing a structure ID to view an overview of hydrogen bonds, hydrophobic contacts, salt bridges, and key sites, with downloadable figures or reports. For batch analysis, provide multiple structures. The interaction types available and result formats are subject to the page display at submission time.
When submitting, explain the receptor, ligand, and pocket_file respectively, and provide chains, key residues, reference ligands, or pocket boundaries. For specific wording, see Ask questions and upload files.
Protein properties
Suited to computing or organizing the physicochemical properties, structural features, functional annotations, pathways, and interaction networks of proteins.
| Analysis direction | Representative resources or tools |
|---|---|
| Molecular weight, isoelectric point, extinction coefficient, amino acid composition, hydrophobicity | Protein property calculators |
| pKa calculation | PROPKA |
| Secondary structure, solvent-accessible surface area, hydrogen bond networks | DSSP |
| Secondary structure prediction | PSIPRED |
| Functional site or property prediction | VenusG, VenusX (submit a sequence; returns a residue site list or property scores) |
| Protein interaction networks and pathway enrichment | STRING |
| Functional annotation and pathway enrichment | GO, KEGG, UniProt |
| Stability assessment and molecular dynamics trajectory analysis | GROMACS |
| Quality screening of generated or engineered sequences | Naturalness scoring models |
Provide the sequence or structure, the properties of interest, experimental conditions (such as pH, temperature, or ionic strength), and the metrics or charts you want returned. If using VenusG or VenusX for functional site or property prediction, confirm the prediction type and sequence information before submitting, then review the returned residue site list or property scores; actual availability, input requirements, and result formats are subject to the prompts shown at submission.
Question-writing tips
Whichever domain you choose, state first the subject, task goal, existing data, constraints, output format, and evidence requirements. If unsure about the specific tool, simply describe the research question you want completed. For complex tasks, the agent may first present a research plan or report outline and wait for your confirmation; before confirming, you can still add subjects, goals, constraints, and output requirements. For how to organize questions, upload files, and confirm docking inputs, see Ask questions and upload files.