IMPRESS

Integrated Machine-learning for PRotEin Structures at Scale

IMPRESS is an asynchronous framework for managing complex protein design pipelines with adaptive decision-making capabilities. It is built for deploying heterogeneous scientific worflows (mixed CPU/GPU and data sharing) in high-performance computing environments. Using a building-block approach to workflow design, IMPRESS enables high-throughput campaigns based on foundation models like AlphaFold and ESM2 or with custom models requiring runtime training and optimization.

Features

🧬	Protein Design Pipelines: Prebuilt and custom workflows
🔄	Adaptive Execution: Dynamic pipeline spawning
⚡	HPC Optimized: High-performance async execution
🎯	Flexible Architecture: Standard and user-defined pipelines

Quick Example

from impress import ImpressBasePipeline, ImpressManager

class MyPipeline(ImpressBasePipeline):
    def register_pipeline_tasks(self):
        @self.auto_register_task()
        async def analyze(self):
            return "echo 'Analyzing sequences'"

    async def run(self):
        await self.analyze()
        await self.invoke_adaptive_step(wait=False)

async def run_dummy_pipelines():

    manager = ImpressManager(execution_backend=ThreadExecutionBackend({}))

    await manager.start(pipeline_setups=[{'name': 'p1', 'config': {}, 
                                          'type': MyPipeline},
                                         {'name': 'p2', 'config': {}, 
                                          'type': MyPipeline},
                                         {'name': 'p3', 'config': {},
                                          'type': MyPipeline}])

asyncio.run(run_dummy_pipelines())