We curate extensive resources encompassing models, datasets and evaluations for Scientific Large Language Models (Sci-LLMs) , with a specific focus on the domains of biology and chemistry. We primarily emphasizes scientific languages, including textual, molecular, protein, and genomic languages, as well as their multimodal combinations. See  Scientific Large Language Models: A Survey on Biological & Chemical Domains for more details! We enumerate available resources for Sci-LLMs, open-source and maintain the related materials at GitHub, thereby facilitating accessibility for newcomers to the field.

We provide a systematic review of existing Sci-LLMs, including

• Textual Scientific Large Language Models (Text-Sci-LLMs). Text-Sci-LLMs are a category of LLMs that are specifically trained on vast amounts of textual scientific data. These models excel in understanding, generating, and interacting with human language in its written form.
• Molecular Large Language Models (Mol-LLMs). Mol-LLMs are specialized LLMs trained on molecular data, allowing them to understand and predict chemical properties and behaviors of molecules. This distinctive expertise renders them invaluable tools in drug discovery, materials science, and understanding complex chemical interactions.
• Protein Large Language Models (Prot-LLMs). Prot-LLMs are trained specifically on protein-related data, including amino acid sequences, protein folding patterns, and other biological data relevant to proteins. Consequently, they possess the capability to accurately predict protein structures, functions, and interactions.
• Genomic Large Language Models (Gene-LLMs). Gene-LLMs, with a primary focus on genomic data, are trained to understand and predict aspects of genetics and genomics. They can be used to analyze DNA sequences, understand genetic variations, and assist in genetic research endeavors such as identifying genetic markers for diseases or understanding evolutionary biology.
• Multimodal Scientific Large Language Models (MM-Sci-LLMs). MM-Sci-LLMs are the most advanced and versatile models, capable of processing and integrating multiple types of scientific data. They can handle text, molecules, proteins, and more, making them suitable for complex scientific research that spans different data types, making them invaluable in interdisciplinary research areas.

Last updated: Jun 30, 2024

Last updated: Jun 30, 2024