openapi: 3.1.0
info:
  title: NVIDIA NIM Biology (BioNeMo) API
  description: >
    NVIDIA BioNeMo NIMs for drug discovery and structural biology. Each model is
    a containerized microservice with its own task-specific payload but a
    consistent JSON contract. Includes protein structure prediction
    (AlphaFold2, ESMFold, OpenFold), protein generation (ProtGPT2,
    RFDiffusion), molecular property prediction (MolMIM), small molecule
    generation, and molecular docking (DiffDock).
  version: '2026-05-25'
  contact:
    name: NVIDIA Developer Support
    url: https://forums.developer.nvidia.com/c/ai-data-science/nemo-llm-service/
  license:
    name: NVIDIA AI Enterprise License
    url: https://www.nvidia.com/en-us/data-center/products/ai-enterprise/
servers:
  - url: https://integrate.api.nvidia.com
    description: NVIDIA-hosted NIM endpoint
  - url: http://localhost:8000
    description: Self-hosted NIM container default
security:
  - BearerAuth: []
tags:
  - name: Biology
    description: Protein, molecule, and structural-biology operations
paths:
  /v1/biology/nvidia/alphafold2/predict-structure-from-sequence:
    post:
      summary: Predict Protein Structure
      description: Predict the 3D structure of a protein from its amino acid sequence using AlphaFold2 NIM.
      operationId: predictProteinStructure
      tags:
        - Biology
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              required: [sequence]
              properties:
                sequence:
                  type: string
                  description: Single-letter amino acid sequence.
                databases:
                  type: array
                  items:
                    type: string
                  description: MSA databases to use (e.g. `uniref90`, `mgnify`, `small_bfd`).
                relax_prediction:
                  type: boolean
                  default: true
      responses:
        '200':
          description: Predicted structure.
          content:
            application/json:
              schema:
                type: object
                properties:
                  pdbs:
                    type: array
                    items:
                      type: string
                      description: PDB-format structure.
        '400':
          description: Invalid sequence.
        '401':
          description: Missing or invalid API key.
  /v1/biology/mit/diffdock:
    post:
      summary: Dock Ligand To Protein
      description: Predict the binding pose of a small molecule ligand against a protein using DiffDock NIM.
      operationId: dockLigand
      tags:
        - Biology
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              required: [protein, ligand]
              properties:
                protein:
                  type: string
                  description: PDB-format protein structure.
                ligand:
                  type: string
                  description: SDF or SMILES ligand representation.
                num_poses:
                  type: integer
                  default: 10
                time_divisions:
                  type: integer
                  default: 20
                steps:
                  type: integer
                  default: 18
      responses:
        '200':
          description: Predicted binding poses.
          content:
            application/json:
              schema:
                type: object
                properties:
                  ligand_positions:
                    type: array
                    items:
                      type: string
                  position_confidence:
                    type: array
                    items:
                      type: number
        '400':
          description: Invalid input.
  /v1/biology/nvidia/molmim/generate:
    post:
      summary: Generate Small Molecules
      description: Generate small molecule SMILES around a seed molecule using NVIDIA MolMIM NIM.
      operationId: generateMolecules
      tags:
        - Biology
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              required: [smi]
              properties:
                smi:
                  type: string
                  description: Seed SMILES string.
                algorithm:
                  type: string
                  enum: [CMA-ES]
                num_molecules:
                  type: integer
                  default: 10
                property_name:
                  type: string
                  enum: [QED, plogP]
                minimize:
                  type: boolean
                min_similarity:
                  type: number
                  default: 0.3
                particles:
                  type: integer
                  default: 30
                iterations:
                  type: integer
                  default: 10
      responses:
        '200':
          description: Generated molecules.
          content:
            application/json:
              schema:
                type: object
                properties:
                  generated:
                    type: array
                    items:
                      type: object
                      properties:
                        sample:
                          type: string
                        score:
                          type: number
components:
  securitySchemes:
    BearerAuth:
      type: http
      scheme: bearer
      bearerFormat: nvapi-...