Abstract:Background: As large language models continue to expand in size and diversity, their substantial potential and the relevance of their applications are increasingly being acknowledged. The rapid advancement of these models also holds profound implications for the long-term design of stimulus-responsive materials used in drug delivery. Methods: The large model used Hugging Face's Transformers package with BigBird, Gemma, and GPT NeoX architectures. Pre-training used the PubChem dataset, and fine-tuning used QM7b. Chemist instruction training was based on Direct Preference Optimization. Drug Likeness, Synthetic Accessibility, and PageRank Scores were used to filter molecules. All computational chemistry simulations were performed using ORCA and Time-Dependent Density-Functional Theory. Results: To optimize large models for extensive dataset processing and comprehensive learning akin to a chemist's intuition, the integration of deeper chemical insights is imperative. Our study initially compared the performance of BigBird, Gemma, GPT NeoX, and others, specifically focusing on the design of photoresponsive drug delivery molecules. We gathered excitation energy data through computational chemistry tools and further investigated light-driven isomerization reactions as a critical mechanism in drug delivery. Additionally, we explored the effectiveness of incorporating human feedback into reinforcement learning to imbue large models with chemical intuition, enhancing their understanding of relationships involving -N=N- groups in the photoisomerization transitions of photoresponsive molecules. Conclusions: We implemented an efficient design process based on structural knowledge and data, driven by large language model technology, to obtain a candidate dataset of specific photoswitchable molecules. However, the lack of specialized domain datasets remains a challenge for maximizing model performance.

Large Language Models Open New Way of AI-Assisted Molecule Design for Chemists

ChatMol: Interactive Molecular Discovery with Natural Language

Large Language Models as Molecular Design Engines

Interactive Molecular Discovery with Natural Language

Language models in molecular discovery

Knowledge-informed molecular design for zeolite synthesis using ChatGPT towards human-machine collaboration

An Autonomous Large Language Model Agent for Chemical Literature Data Mining

Small Molecule Optimization with Large Language Models

Synergizing Human Expertise and AI Efficiency with Language Model for Microscopy Operation and Automated Experiment Design

Chemist-X: Large Language Model-empowered Agent for Reaction Condition Recommendation in Chemical Synthesis

MatChat: A Large Language Model and Application Service Platform for Materials Science

Generative AI-Driven Molecular Design: Combining Predictive Models and Reinforcement Learning for Tailored Molecule Generation

De novo drug design as GPT language modeling: large chemistry models with supervised and reinforcement learning

Knowledge-informed generation of organic structure-directing agents for zeolites using ChatGPT towards human-machine collaborative molecular design

Crossing New Frontiers: Knowledge-Augmented Large Language Model Prompting for Zero-Shot Text-Based De Novo Molecule Design

DrugAssist: A Large Language Model for Molecule Optimization

Unlocking comprehensive molecular design across all scenarios with large language model and unordered chemical language

SynAsk: Unleashing the Power of Large Language Models in Organic Synthesis

Discovering Photoswitchable Molecules for Drug Delivery with Large Language Models and Chemist Instruction Training

A Review of Large Language Models and Autonomous Agents in Chemistry

A multi-agent-driven robotic AI chemist enabling autonomous chemical research on demand