AutoProteinEngine: LLM-Driven AutoML for Protein Engineering

Protein engineering, while crucial for biomedical advancements, often relies on inefficient and resource-intensive conventional methods. Although deep learning (DL) models offer significant promise, their implementation remains challenging for biologists lacking specialized computational skills. This gap hinders the broader adoption of advanced DL techniques in the field. To bridge this divide, AutoProteinEngine (AutoPE) has been developed as an agent framework leveraging large language models (LLMs) for multimodal automated machine learning (AutoML) in protein engineering. AutoPE uniquely enables biologists to interact with DL models using natural language, thereby lowering the entry barrier. It integrates LLMs with AutoML for tasks such as model selection across protein sequence and graph modalities, automatic hyperparameter optimization, and automated data retrieval from protein databases, demonstrating substantial performance improvements in real-world applications.