Andrew Shen | Unlocking LLM Creativity in Science through Analogical Reasoning

Abstract

Autonomous science promises to augment scientific discovery, particularly in complex fields like biomedicine. However, this requires AI systems that can consistently generate novel and diverse solutions to open-ended problems. We evaluate LLMs on the task of open-ended solution generation and quantify their tendency to mode collapse into low-diversity generations. To mitigate this mode collapse, we introduce analogical reasoning (AR) as a new approach to solution generation. AR generates analogies to cross-domain problems based on shared relational structure, then uses those analogies to search for novel solutions. Compared to baselines, AR discovers significantly more diverse generations (improving solution diversity metrics by 90-173%), generates novel solutions over 50% of the time (compared to as little as 1.6% for baselines), and produces high-quality analogies. To validate the real-world feasibility of AR, we implement AR-generated solutions across four biomedical problems, yielding consistent quantitative gains. AR-generated approaches achieve a nearly 13-fold improvement on distributional metrics for perturbation effect prediction, outperform all baselines on AUPRC when predicting cell-cell communication, infer brain region interactions with a high Spearman correlation (ρ=0.729) to published methods, and establish state-of-the-art performance on 2 datasets for oligonucleotide property prediction. The novel and diverse solutions produced by AR can be used to augment the search space of existing solution generation methods.

Bio 

Andrew is a second-year PhD student in Biomedical Data Science at Stanford, advised by James Zou. He is broadly interested in problems at the intersection of AI and science, and his current work focuses on autonomous scientific discovery. Before Stanford, he was a research associate with Marinka Zitnik at Harvard Medical School, where he developed machine learning methods for medicine and science, including work on evolutionary reasoning of protein language models and molecular AI generalizability. He completed a MS in Artificial Intelligence from Northwestern University and a BS in Bioinformatics from UC San Diego.

About the Center for Neural Data Science Seminar Series 

The Center for Neural Data Science Seminar Series is a platform for trainees across campus to share  insights and innovative approaches that bridge the gap between neuroscience and data science. 

As neuroscience continues to generate vast amounts of data—from intricate neural circuit maps to large-scale brain activity recordings—the need for interdisciplinary expertise in data science, statistics, and engineering has never been more critical. 

The Center for Neural Data Science mission is to advance brain research through the development of cutting-edge analytical methodologies and collaborative approaches. Stanford's affiliates are invited to join this vibrant community dedicated to transformative discoveries.

This seminar series is only offered in person. 

Happy Hour/Reception for Attendees: 

Seminars attendees are invited to stay for a happy hour reception after the seminar to continue the conversation with the speaker.

Sign up to learn about all our upcoming events