A survey on large language models in biology and chemistry

2026-04-08 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Life Sciences & Biology, Physical Sciences & Chemistry, Health & Medical Research · Depth: Expert, extended

Summary

A 2026 review article in "Experimental & Molecular Medicine" surveys the application of large language models (LLMs) in biology and chemistry, highlighting their evolution from molecular representation to generation and optimization. The review details key molecular representation strategies for biological macromolecules (e.g., protein/nucleotide sequences, single-cell data) and small organic compounds (e.g., SMILES strings, graph-based encodings, 3D point clouds). It covers core model architectures like BERT-like encoders, GPT-like decoders, and encoder-decoder transformers, along with pretraining strategies such as self-supervised learning, multitask learning, and retrieval-augmented generation. Key biomedical applications include protein structure prediction, de novo protein/molecular design, genomic analysis, and reaction prediction. The article also explores emerging agentic and interactive AI systems for automating scientific discovery, while addressing technical, ethical, and regulatory considerations.

Key takeaway

For AI Scientists and Research Scientists developing computational tools in biomedicine, understanding the nuanced interplay between molecular representation, model architecture, and training strategies is crucial. You should prioritize developing robust tokenization and multimodal integration techniques to effectively translate complex biological and chemical data into formats LLMs can process, thereby accelerating discovery and design cycles. Consider adopting agentic AI systems to automate experimental design and hypothesis generation.

Key insights

LLMs are transforming molecular sciences by treating biological and chemical systems as structured languages.

Principles

• Representational design is fundamental for scientific LLMs.
• Model architecture and training data scale are critical.
• Multimodal frameworks enhance understanding of biological systems.

Method

LLMs adapt to scientific domains by converting complex molecular information into processable formats, utilizing diverse architectures and pretraining strategies like self-supervised and multitask learning.

In practice

• Use prompt engineering for rapid, low-resource LLM adaptation.
• Fine-tune LLMs on domain-specific corpora for task specificity.
• Integrate LLMs with agentic systems for automated workflows.

Topics

• Large Language Models
• Bio/Chemical Language Models
• Molecular Representation
• Protein Structure Prediction
• De Novo Molecular Design

Best for: AI Scientist, Research Scientist, Director of AI/ML

Related on AIssential

• Rethinking Molecular Text Representations for LLMs: An Empirical Study — LLM molecular competence is strongly representation-dependent, with no single representation optimal across all tasks.
• IEEE Transactions on Artificial Intelligence, Volume 7, Issue 2, February 2026 — The volume showcases diverse AI advancements, from LLM applications in autonomous driving to specialized models for protein language processing.
• A multimodal large language model for materials science — MatterChat unifies material structure and text using a modular LLM, improving property prediction and scientific reasoning in materials science.
• Building LLM From Scratch: Understanding How Large Language Models Work — Large Language Models fundamentally predict the next token by processing numerical representations of text through attention mechanisms.
• A family of large language models for materials research with insights into model adaptability in continued pretraining — Domain-adapted LLMs significantly enhance materials research, but overtraining can hinder further specialization.
• How Large Language Models Actually Work (Explained Simply) — LLMs are statistical prediction engines, not reasoning entities, whose capabilities emerge from massive scale and sophisticated attention mechanisms.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.