03 Core concepts and features
Core skill examples
Computational workflows
ReactNet + ReactGen: reaction prediction and reaction network generation
- Combines reaction enumeration, transition-state searches with rapid ReactOT generation, and IRC validation.
- Choose methods and accuracy levels to search multiple reactions automatically in parallel and build a systematic picture of the reaction energy network.
- Enter the following prompt in the chat input:
- “/reactnet calculate the energy barrier for [reaction] ([reactants] → [products])”
- “/reactnet calculate a reaction network with [X] layers for [reactant]”
Latest example: https://mira.deepprinciple.com/share/sessions/M-qebmNF
Reactify: quantum chemistry engine
- Quantum chemistry at multiple accuracy levels, with a complete XTB / CREST / DFT method suite and packaged workflows for consistent, reproducible execution.
- Automates modeling, calculations, and post-processing with cloud GPU acceleration and parallel batch processing of multiple molecules.
- Enter “/reactify calculate [task]” in the chat input. The video below shows Reactify reproducing a published redox-potential calculation, with results consistent with experimental values.
Latest example: https://mira.deepprinciple.com/share/sessions/Eg0D1-oG
ReactNavi: retrosynthesis design
- Combines template-based, neural-network, and hybrid methods for retrosynthesis design.
- Set constraints such as available starting materials and reaction conditions. The Agent adjusts search depth and expansion strategy to the molecule's complexity, returning multiple routes and visualizations.
- Enter “/reactnavi design a retrosynthetic route for [molecule]” in the chat input.
Latest example: https://mira.deepprinciple.com/share/sessions/V-Cd664-
MPA: material property prediction model
- Predict 42 molecular properties: boiling, melting, and flash points; critical parameters (Tc / Pc / Vc); liquid density; vapor pressure; gas and liquid viscosity; thermal conductivity; heat capacity; enthalpy of formation and Gibbs free energy; heat of combustion; water solubility; logD; surface tension; refractive index; dielectric constant; dipole moment; explosive limits (LEL / UEL); autoignition temperature; and more. One call can predict multiple properties for a batch of molecules.
- Provide input without manual modeling: enter SMILES strings, or molecule names for automatic resolution. The service generates 3D conformers, extracts features, runs MPA inference, and returns values in their original units, with automatic parallel batch processing.
- Enter “/material-property-prediction predict [properties] of [molecules]” in the chat input.
Latest example: https://mira.deepprinciple.com/share/sessions/sod-Vyxr
MDHub: molecular dynamics engine
- Molecular dynamics across multiple systems and sampling levels, using LAMMPS, OpenFF, and Packmol workflows. Supports small-molecule liquids, electrolytes, water, and selected linear polymers, with analyses of density, radial distribution functions, chain dimensions, and dynamics.
- After you confirm the model, Mira connects modeling, parallel cloud calculations, and post-processing. Deliverables include input files, property results, charts, trajectories, and reports. If a run fails, Mira can diagnose logs, correct the issue, and retry.
- Enter “/mdhub calculate [task]” in the chat input. The video below shows a classic polyethylene melt simulation, including model review, packing corrections, LAMMPS calculations, and an initial comparison with published results.
Latest example: https://mira.deepprinciple.com/share/sessions/2gcmfmI2
General research workflows
Deep academic research
- Iterative search and evidence checking: builds a search strategy from your research question, searches broadly for candidate papers, retrieves full texts for key evidence, and fills gaps through additional searches.
- Structured evidence and traceability: key conclusions include source links, confidence assessments, and limitations. Conflicting evidence is presented explicitly.
- Produces a structured Markdown research report with numbered inline citations and clickable references, covering the question, key findings, limitations, and open questions.
- Enter “/deep-research research [topic]” in the chat input.
Broad patent search
- Automatic queries: generates four progressively broader search queries from a technical requirement, using Google Patents advanced syntax, including field filters and proximity operators.
- Filtering and analysis: filter by legal status, jurisdiction, date range, and applicant, with sorting and pagination.
- Batch full-text download: select patent numbers and download their PDFs.
- Enter “/patent-search research patents for [topic]” in the chat input.
Research brainstorming
- Structured research decisions: provide your research goal and available resources. Mira guides you through five steps: clarify the goal → identify variables → explore technical approaches → evaluate against shared criteria → recommend a preferred approach. Covers topic selection, experimental design, route comparison, parameter optimization, and paper planning across chemistry, biology, medicine, computer science, and social science.
- Route comparison and visual decision boards: each candidate includes its mechanism, resource needs, key variables, largest risk, and fastest validation method. Mira can generate an offline HTML board with comparison cards, parameter sliders for exploring effects, and a validation-test matrix.
- Enter
/brainstormingin the chat input.
More specialized skill packages:
Battery simulation: coupled electrochemical and thermal simulation
- 15 battery parameter sets covering common NMC / NCA / LFP / LCO cathodes and graphite, silicon–carbon, and SiOx anodes for commercial cells including cylindrical and pouch formats.
- P2D / DFN: uses Newman's porous-electrode theory to solve ion transport, potential distributions, and thermal effects from electrochemical principles, for mechanism studies, high-rate operation, and cycling degradation analysis.
- ECM: fast equivalent-circuit models fitted to HPPC experimental data, for BMS control prototyping.
- PyBOP: uses Bayesian and gradient-based optimization to infer model parameters from measured charge–discharge curves, supporting calibration when experimental data is available but parameters are unknown.
- Enter “/battery-simulation” in the chat input.
Latest example: https://mira.deepprinciple.com/share/sessions/AQnJ47b0
Bioactive molecule design for nutrition and personal care: protein docking and binding-mode analysis
- End-to-end folding prediction: predict 3D structures from amino-acid sequences. Supports protein complexes, protein–DNA / RNA / ligand complexes, and covalently modified systems, from single chains to complex all-atom inputs.
- Submit asynchronous cloud calculations: automates modeling, inference, and result downloads. Returns CIF / PDB structures, PAE matrices, residue-distance maps, interchain ipTM, and confidence metrics, with sampling and automatic ranking of multiple candidate structures.
- Enter “/esmfold2 ” in the chat input.
Latest example: https://mira.deepprinciple.com/share/sessions/FBaLcs7o