Tag
This paper benchmarks pretrained molecular language models for virtual screening and demonstrates that explicit domain adaptation improves their performance and sample efficiency across drug discovery, materials, and catalysis libraries.