Beyond Chemistry Knowledge: Embedding Pedagogical Content Knowledge through an AI-Augmented Retrosynthesis Program for Chemistry Education
- 1Peking University , , ,
Abstract
Synthetic route planning is a core learning objective in university organic chemistry, yet a challenging one requiring the development of advanced problem-solving skills. Existing computational tools often propose routes beyond the intended curriculum, limiting their suitability as learning resources. In this work, we introduce Retro-Learn, an educational retrosynthesis platform that integrates a curriculum-aligned reaction template library with large language model interfaces. The platform incorporates selected aspects of pedagogical content knowledge by aligning available reactions with the scope and sequence of instruction. An agent skill and a web-based chatbot enable route generation and visualization through natural language interaction. Evaluation using cumulative chapter-specific template libraries showed how the addition of reaction knowledge enables new routes and shorter solutions. Blind assessments of undergraduate examination problems showed higher mean chemical correctness scores than those of database-derived routes and scores comparable to or higher than those of general-purpose language models. Student survey responses indicated a plurality preference for Retro-Learn over general-purpose tools as a synthesis-planning learning aid. These findings illustrate how curriculum-specific chemical informatics and language model interfaces can translate pedagogical considerations into computational support for organic chemistry learning.
Related
- RetroMol: Knowledge Distillation of Energy-Based Retrosynthesis Models
- Digitalizing Manufacturing Chemistry Targets: A Benchmark for Retrosynthesis Tools
- GPU infrastructure for ML/MM with electrostatic embedding for enzyme chemistry
- Quantum-Classical Effective Fragment Potential Embedding for Condensed-Phase Quantum Chemistry
- Functional-Group Knowledge Is Structured in Language Models Before It Is Spoken
BibTeX
@article{wang2026beyond,
title = {{Beyond Chemistry Knowledge: Embedding Pedagogical Content Knowledge through an AI-Augmented Retrosynthesis Program for Chemistry Education}},
author = {Zihan Wang and Changyu Hu and Junren Li and Yue Su and Mingyuan Lu and Hua Lu and Luhua Lai and Xiao Li},
journal = {JACS Au},
year = {2026},
month = oct,
doi = {10.1021/jacsau.6c01198},
url = {https://pubs.acs.org/jaaucr/article/doi/10.1021/jacsau.6c01198/5455974/Beyond-Chemistry-Knowledge-Embedding-Pedagogical},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Beyond Chemistry Knowledge: Embedding Pedagogical Content Knowledge through an AI-Augmented Retrosynthesis Program for Chemistry Education
AU - Zihan Wang
AU - Changyu Hu
AU - Junren Li
AU - Yue Su
AU - Mingyuan Lu
AU - Hua Lu
AU - Luhua Lai
AU - Xiao Li
PY - 2026
DA - 2026/10/22
JO - JACS Au
DO - 10.1021/jacsau.6c01198
UR - https://pubs.acs.org/jaaucr/article/doi/10.1021/jacsau.6c01198/5455974/Beyond-Chemistry-Knowledge-Embedding-Pedagogical
AB - Synthetic route planning is a core learning objective in university organic chemistry, yet a challenging one requiring the development of advanced problem-solving skills. Existing computational tools often propose routes beyond the intended curriculum, limiting their suitability as learning resources. In this work, we introduce Retro-Learn, an educational retrosynthesis platform that integrates a curriculum-aligned reaction template library with large language model interfaces. The platform incorporates selected aspects of pedagogical content knowledge by aligning available reactions with the scope and sequence of instruction. An agent skill and a web-based chatbot enable route generation and visualization through natural language interaction. Evaluation using cumulative chapter-specific template libraries showed how the addition of reaction knowledge enables new routes and shorter solutions. Blind assessments of undergraduate examination problems showed higher mean chemical correctness scores than those of database-derived routes and scores comparable to or higher than those of general-purpose language models. Student survey responses indicated a plurality preference for Retro-Learn over general-purpose tools as a synthesis-planning learning aid. These findings illustrate how curriculum-specific chemical informatics and language model interfaces can translate pedagogical considerations into computational support for organic chemistry learning.
SN - 2691-3704
ER -