EquiLLM integrates Large Language Models with geometric equivariance, enabling accurate 3D structure prediction while leveraging broader knowledge.

• Addresses a fundamental limitation of traditional Graph Neural Networks by incorporating language understanding
• Maintains critical E(3)-equivariance properties for spatial reasoning
• Demonstrates improved performance in molecular dynamics and structural prediction tasks
• Shows particular promise for antibody design in medical applications

This research represents a significant advancement for medical research, enabling more accurate predictions of complex biological structures such as antibodies—a critical component in drug discovery and therapeutic development.

Original Paper: Large Language-Geometry Model: When LLM meets Equivariance