Navigating Chemical-Linguistic Sharing Space with Heterogeneous Molecular Encoding

Hao Li, Liuzhenghao Lv, Yu Wang, Zijun Chen, Yuyang Liu, Li Yuan†, Yonghong Tian†

Project

• Official Homepage for MHE, a Nature Communications work on heterogeneous molecular encoding for chemical language models. MHE bridges molecular structures and natural language by unifying sequence, topology, geometry, and fragment-level information, supporting bidirectional molecular understanding and generation across chemical-linguistic space.

Beyond Chemical QA: Evaluating LLM’s Chemical Reasoning with Modular Chemical Operations

Hao Li, He Cao, Bin Feng, Yanjun Shao, Xiangru Tang, Zhiyuan Yan, Li Yuan†, Yonghong Tian†, Yu Li†

Project

• Official Homepage for the ChemCoTBench, the first step-wise reasoning benchmark for LLMs in the Chemical domain, covering molecular understanding, editing, optimization, and reaction-related predictions. We have contributed an expert-annotated benchmark dataset of 1.5K samples, along with a high-quality SFT-RL reasoning dataset comprising 22K instances.

Hierarchical banzhaf interaction for general video-language representation learning

Hao Li, Peng Jin, Shuicheng Yan, Li Yuan†, Jie Chen†

Project

• Official Homepage for the HBI, a new approach that models video-text as game players using multivariate cooperative game theory to handle uncertainty during fine-grained semantic interactions with diverse granularity, flexible combination, and vague intensity.

FreestyleRet: Retrieving Images from Style-Diversified Queries

Hao Li, Yanhao Jia, Peng Jin, Zesen Cheng, Kehan Li, Jialu Sui, Chang Liu, Li Yuan

Project

• Official Code for the FreestyleRet framework. Official Release for the Diversified-Style Retrieval Dataset (DSR).

Decoupled peak property learning for efficient and interpretable ECD spectra prediction

Hao Li, Da Long, Li Yuan, Yu Wang, Yonghong Tian, Xinchang Wang, Fanyang Mo

Project

• Existing predictive approaches lack the consideration of ECD spectra due to the data scarcity, and the interpretability to achieve trust-worthy prediction. Here, we establish a large-scale dataset for Chiral Molecular ECD spectra (CMCDS) and propose ECDFormer for accurate and interpretable ECD spectra prediction. ECDFormer decomposes ECD spectra into peak entities, employs the QFormer architecture to learn peak properties, and renders peaks into spectra. Compared to spectra sequence prediction methods, our decoupled peak prediction approach substantially enhances both accuracy and efficiency, improving the peak symbol accuracy from 37.3% to 72.7% and decreasing the time cost from an average of 4.6 CPU hours to 1.5 seconds.

Diffusionret: Generative text-video retrieval with diffusion model

Hao Li, Peng Jin, Zesen Cheng, Kehan Li, Chang Liu, Li Yuan, Jie Chen

Project

• Official Code for the DiffusionRet model. We propose DiffusionRet, a generative text-video retrieval framework that models retrieval as generating joint distributions from noise. Unlike standard methods, it excels not only on standard benchmarks but also in retrieving out-of-distribution videos, demonstrating superior generalization.