Spaces:
Sleeping
Sleeping
| """ | |
| Data models for the PharmaDDI Environment. | |
| The PharmaDDI environment simulates clinical pharmacist drug interaction | |
| checking. An agent receives patient medication lists and must identify | |
| drug-drug interactions, classify severity, and recommend clinical actions. | |
| Database: 60+ drugs, 120+ interactions across all major therapeutic classes. | |
| Grounded in DDI-Bench methodology (Shen et al., Bioinformatics 2025). | |
| """ | |
| from typing import List, Optional, Dict, Any | |
| from openenv.core.env_server.types import Action, Observation | |
| from pydantic import Field, BaseModel | |
| # --------------------------------------------------------------------------- | |
| # Sub-models | |
| # --------------------------------------------------------------------------- | |
| class MedicationInfo(BaseModel): | |
| """A single medication in the patient's list.""" | |
| name: str = Field(..., description="Drug generic name (lowercase)") | |
| therapeutic_class: str = Field(..., description="Therapeutic class (e.g. SSRI, NSAID, statin)") | |
| common_dose: str = Field(..., description="Common dose (e.g. 20mg)") | |
| frequency: str = Field(..., description="Dosing frequency (e.g. once daily)") | |
| class InteractionReport(BaseModel): | |
| """Agent's report of a single drug-drug interaction.""" | |
| drug_a: str = Field(..., description="First drug name (lowercase generic)") | |
| drug_b: str = Field(..., description="Second drug name (lowercase generic)") | |
| severity: str = Field( | |
| ..., | |
| description="Interaction severity: minor | moderate | major | contraindicated" | |
| ) | |
| clinical_effect: str = Field( | |
| default="", | |
| description="Description of the clinical mechanism and consequence" | |
| ) | |
| recommendation: str = Field( | |
| default="", | |
| description="Recommended clinical action: monitor | adjust_dose | substitute | discontinue" | |
| ) | |
| # --------------------------------------------------------------------------- | |
| # OpenEnv Action & Observation | |
| # --------------------------------------------------------------------------- | |
| class PharmaDDIAction(Action): | |
| """ | |
| Agent's response analyzing drug interactions in a patient scenario. | |
| The agent submits a list of detected drug-drug interactions. | |
| Multi-step: the agent can refine its answer over multiple steps | |
| using feedback from the grader. Set done=True when confident. | |
| """ | |
| interactions_found: List[InteractionReport] = Field( | |
| default_factory=list, | |
| description="List of identified drug-drug interaction pairs with clinical details" | |
| ) | |
| done: bool = Field( | |
| default=False, | |
| description="Set True when the agent is confident and wants to end the episode" | |
| ) | |
| class PharmaDDIObservation(Observation): | |
| """ | |
| Patient scenario presented to the agent for drug interaction analysis. | |
| Returned by both reset() and step(). After step(), the feedback field | |
| contains the grader's detailed assessment of the previous submission, | |
| and score reflects the best score achieved so far in this episode. | |
| """ | |
| # --- Task identity --- | |
| task_name: str = Field( | |
| default="", | |
| description="Task identifier: easy_pair_check | medium_multi_drug | hard_polypharmacy" | |
| ) | |
| task_difficulty: str = Field( | |
| default="", | |
| description="Difficulty level: easy | medium | hard" | |
| ) | |
| # --- Patient profile --- | |
| patient_id: str = Field(default="", description="Unique patient identifier for this episode") | |
| patient_age: int = Field(default=0, description="Patient age in years") | |
| patient_conditions: List[str] = Field( | |
| default_factory=list, | |
| description="Patient's active medical conditions (e.g. 'atrial_fibrillation', 'type2_diabetes')" | |
| ) | |
| medications: List[MedicationInfo] = Field( | |
| default_factory=list, | |
| description="Current medication list — agent must check all pairs for interactions" | |
| ) | |
| num_medications: int = Field(default=0, description="Number of medications in the list") | |
| # --- Task guidance --- | |
| instructions: str = Field( | |
| default="", | |
| description="Task-specific clinical instructions for the agent" | |
| ) | |
| # --- Step feedback --- | |
| feedback: str = Field( | |
| default="", | |
| description=( | |
| "Grader feedback from the most recent step. Includes: score breakdown, " | |
| "which pairs were correct, missed interactions, false positives, " | |
| "and mechanism hints for missed pharmacokinetic interactions." | |
| ) | |
| ) | |
| score: float = Field( | |
| default=0.0, | |
| description="Best score achieved so far this episode (0.0–1.0, cumulative best)" | |
| ) | |
| # --- Hints --- | |
| total_interactions: int = Field( | |
| default=-1, | |
| description=( | |
| "Hint: total number of real interactions in this scenario. " | |
| "Provided only for easy_pair_check. " | |
| "-1 means not disclosed (medium and hard tasks)." | |
| ) | |
| ) | |
| actions_remaining: int = Field( | |
| default=10, | |
| description="Number of steps remaining before the episode ends automatically" | |
| ) | |
| # --- Curriculum metadata --- | |
| metadata: Dict[str, Any] = Field( | |
| default_factory=dict, | |
| description=( | |
| "Episode metadata including: episode_id, step count, task name, " | |
| "curriculum_focus (which drug class is being targeted this episode), " | |
| "and curriculum_status (agent performance per drug class category)." | |
| ) | |
| ) | |
| # --- Compatibility (kept for backward compat with older clients) --- | |
| done: bool = Field(default=False, description="Whether the episode has ended") | |
| reward: Optional[float] = Field( | |
| default=None, | |
| description="Step reward (improvement over best score so far). None after reset()." | |
| ) |