Le Song
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Le Song has a total of 164 co-authors. Here are the top co-authors by number of shared publications:
Xingyi Cheng
BioMap Research , Beijing 100190
6 shared publicationsJing Gong
BioMap Research, Palo Alto, CA, USA
5 shared publicationsXin Zeng
BioMap Research, Palo Alto, CA, USA
5 shared publications- Hui Li
BioMap
4 shared publications - Jie Tang
Tsinghua University
4 shared publications - Xuegong Zhang
MOE Key Laboratory of Bioinformatics and Bioinformatics Division of BNRIST, Department of Automation, Tsinghua University , Beijing 100084
4 shared publications
Latest Publications
- Open Access
- 12 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-2
CellMap: WebGPU Visualization of Multi-Scale AI-Generated Cell Biology Data
- 12 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Open AccessCellMap: WebGPU Visualization of Multi-Scale AI-Generated Cell Biology Data
- Open Access
- 4 August 2026
- Published by Springer Nature in npj Digital Medicine
MedKGent: a large language model agent framework for constructing temporally evolving medical knowledge graph
- 4 August 2026
- Published by Springer Nature in npj Digital Medicine
Open AccessMedKGent: a large language model agent framework for constructing temporally evolving medical knowledge graph
- Open Access
MemR$^3$: Memory Retrieval via Reflective Reasoning for LLM Agents
Memory systems have been designed to leverage past experiences in Large Language Model (LLM) agents. However, many deployed memory systems primarily optimize compression and storage, with comparatively less emphasis on explicit, closed-loop control of memory retrieval. From this observation, we build memory retrieval as an autonomous, accurate, and compatible agent system, named MemR$^3$, which has two core mechanisms: 1) a router that selects among retrieve, reflect, and answer actions to optimize answer quality; 2) a global evidence-gap tracker that explicitly renders the answering process transparent and tracks the evidence collection process. This design departs from the standard retrieve-then-answer pipeline by introducing a closed-loop control mechanism that enables autonomous decision-making. Empirical results on the LoCoMo benchmark demonstrate that MemR$^3$ surpasses strong baselines on LLM-as-a-Judge score, and particularly, it improves existing retrievers across four categories with an overall improvement on RAG (+7.29%) and Zep (+1.94%) using GPT-4.1-mini backend, offering a plug-and-play controller for existing memory stores.
Open AccessMemR$^3$: Memory Retrieval via Reflective Reasoning for LLM Agents
Memory systems have been designed to leverage past experiences in Large Language Model (LLM) agents. However, many deployed memory systems primarily optimize compression and storage, with comparatively less emphasis on explicit, closed-loop control of memory retrieval. From this observation, we build memory retrieval as an autonomous, accurate, and compatible agent system, named MemR$^3$, which has two core mechanisms: 1) a router that selects among retrieve, reflect, and answer actions to optimize answer quality; 2) a global evidence-gap tracker that explicitly renders the answering process transparent and tracks the evidence collection process. This design departs from the standard retrieve-then-answer pipeline by introducing a closed-loop control mechanism that enables autonomous decision-making. Empirical results on the LoCoMo benchmark demonstrate that MemR$^3$ surpasses strong baselines on LLM-as-a-Judge score, and particularly, it improves existing retrievers across four categories with an overall improvement on RAG (+7.29%) and Zep (+1.94%) using GPT-4.1-mini backend, offering a plug-and-play controller for existing memory stores.
- Open Access
HD-Prot: A Protein Language Model for Joint Sequence-Structure Modeling with Continuous Structure Tokens
Proteins inherently possess a consistent sequence-structure duality. The abundance of protein sequence data, which can be readily represented as discrete tokens, has driven fruitful developments in protein language models (pLMs). A key remaining challenge, however, is how to effectively integrate continuous structural knowledge into pLMs. Current methods often discretize protein structures to accommodate the language modeling framework, which inevitably results in the loss of fine-grained information and limits the performance potential of multimodal pLMs. In this paper, we argue that such concerns can be circumvented: a sequence-based pLM can be extended to incorporate the structure modality through continuous tokens, i.e., high-fidelity protein structure latents that avoid vector quantization. Specifically, we propose a hybrid diffusion protein language model, HD-Prot, which embeds a continuous-valued diffusion head atop a discrete pLM, enabling seamless operation with both discrete and continuous tokens for joint sequence-structure modeling. It captures inter-token dependencies across modalities through a unified absorbing diffusion process, and estimates per-token distributions via categorical prediction for sequences and continuous diffusion for structures. Extensive empirical results show that HD-Prot achieves competitive performance in unconditional sequence-structure co-generation, motif-scaffolding, protein structure prediction, and inverse folding tasks, performing on par with state-of-the-art multimodal pLMs despite being developed under limited computational resources. It highlights the viability of simultaneously estimating categorical and continuous distributions within a unified language model architecture, offering a promising alternative direction for multimodal pLMs.
Open AccessHD-Prot: A Protein Language Model for Joint Sequence-Structure Modeling with Continuous Structure Tokens
Proteins inherently possess a consistent sequence-structure duality. The abundance of protein sequence data, which can be readily represented as discrete tokens, has driven fruitful developments in protein language models (pLMs). A key remaining challenge, however, is how to effectively integrate continuous structural knowledge into pLMs. Current methods often discretize protein structures to accommodate the language modeling framework, which inevitably results in the loss of fine-grained information and limits the performance potential of multimodal pLMs. In this paper, we argue that such concerns can be circumvented: a sequence-based pLM can be extended to incorporate the structure modality through continuous tokens, i.e., high-fidelity protein structure latents that avoid vector quantization. Specifically, we propose a hybrid diffusion protein language model, HD-Prot, which embeds a continuous-valued diffusion head atop a discrete pLM, enabling seamless operation with both discrete and continuous tokens for joint sequence-structure modeling. It captures inter-token dependencies across modalities through a unified absorbing diffusion process, and estimates per-token distributions via categorical prediction for sequences and continuous diffusion for structures. Extensive empirical results show that HD-Prot achieves competitive performance in unconditional sequence-structure co-generation, motif-scaffolding, protein structure prediction, and inverse folding tasks, performing on par with state-of-the-art multimodal pLMs despite being developed under limited computational resources. It highlights the viability of simultaneously estimating categorical and continuous distributions within a unified language model architecture, offering a promising alternative direction for multimodal pLMs.
- Open Access
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 2967-2976
Full-Atom Protein-Protein Interaction Prediction via Atomic Equivariant Attention Network
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 2967-2976
Open AccessFull-Atom Protein-Protein Interaction Prediction via Atomic Equivariant Attention Network
- Open Access
- 2 September 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 8210
Prediction of cellular morphology changes under perturbations with a transcriptome-guided diffusion model
Investigating cell morphology changes after perturbations using high-throughput image-based profiling is increasingly important for phenotypic drug discovery, including predicting mechanisms of action (MOA) and compound bioactivity. The vast space of chemical and genetic perturbations makes it impractical to explore all possibilities using conventional methods. Here we propose MorphDiff, a transcriptome-guided latent diffusion model that simulates high-fidelity cell morphological responses to perturbations. We demonstrate MorphDiff’s effectiveness on three large-scale datasets, including two drug perturbation and one genetic perturbation dataset, covering thousands of perturbations. Extensive benchmarking shows MorphDiff accurately predicts cell morphological changes under unseen perturbations. Additionally, MorphDiff enhances MOA retrieval, achieving an accuracy comparable to ground-truth morphology and outperforming baseline methods by 16.9% and 8.0%, respectively. This work highlights MorphDiff’s potential to accelerate phenotypic screening and improve MOA identification, making it a powerful tool in drug discovery.
- 2 September 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 8210
Open AccessPrediction of cellular morphology changes under perturbations with a transcriptome-guided diffusion model
Investigating cell morphology changes after perturbations using high-throughput image-based profiling is increasingly important for phenotypic drug discovery, including predicting mechanisms of action (MOA) and compound bioactivity. The vast space of chemical and genetic perturbations makes it impractical to explore all possibilities using conventional methods. Here we propose MorphDiff, a transcriptome-guided latent diffusion model that simulates high-fidelity cell morphological responses to perturbations. We demonstrate MorphDiff’s effectiveness on three large-scale datasets, including two drug perturbation and one genetic perturbation dataset, covering thousands of perturbations. Extensive benchmarking shows MorphDiff accurately predicts cell morphological changes under unseen perturbations. Additionally, MorphDiff enhances MOA retrieval, achieving an accuracy comparable to ground-truth morphology and outperforming baseline methods by 16.9% and 8.0%, respectively. This work highlights MorphDiff’s potential to accelerate phenotypic screening and improve MOA identification, making it a powerful tool in drug discovery.
- Open Access
MedKGent: A Large Language Model Agent Framework for Constructing Temporally Evolving Medical Knowledge Graph
The rapid expansion of medical literature presents growing challenges for structuring and integrating domain knowledge at scale. Knowledge Graphs (KGs) offer a promising solution by enabling efficient retrieval, automated reasoning, and knowledge discovery. However, current KG construction methods often rely on supervised pipelines with limited generalizability or naively aggregate outputs from Large Language Models (LLMs), treating biomedical corpora as static and ignoring the temporal dynamics and contextual uncertainty of evolving knowledge. To address these limitations, we introduce MedKGent, a LLM agent framework for constructing temporally evolving medical KGs. Leveraging over 10 million PubMed abstracts published between 1975 and 2023, we simulate the emergence of biomedical knowledge via a fine-grained daily time series. MedKGent incrementally builds the KG in a day-by-day manner using two specialized agents powered by the Qwen2.5-32B-Instruct model. The Extractor Agent identifies knowledge triples and assigns confidence scores via sampling-based estimation, which are used to filter low-confidence extractions and inform downstream processing. The Constructor Agent incrementally integrates the retained triples into a temporally evolving graph, guided by confidence scores and timestamps to reinforce recurring knowledge and resolve conflicts. The resulting KG contains 156,275 entities and 2,971,384 relational triples. Quality assessments by two SOTA LLMs and three domain experts demonstrate an accuracy approaching 90%, with strong inter-rater agreement. To evaluate downstream utility, we conduct RAG across seven medical question answering benchmarks using five leading LLMs, consistently observing significant improvements over non-augmented baselines. Case studies further demonstrate the KG's value in literature-based drug repurposing via confidence-aware causal inference.
Open AccessMedKGent: A Large Language Model Agent Framework for Constructing Temporally Evolving Medical Knowledge Graph
The rapid expansion of medical literature presents growing challenges for structuring and integrating domain knowledge at scale. Knowledge Graphs (KGs) offer a promising solution by enabling efficient retrieval, automated reasoning, and knowledge discovery. However, current KG construction methods often rely on supervised pipelines with limited generalizability or naively aggregate outputs from Large Language Models (LLMs), treating biomedical corpora as static and ignoring the temporal dynamics and contextual uncertainty of evolving knowledge. To address these limitations, we introduce MedKGent, a LLM agent framework for constructing temporally evolving medical KGs. Leveraging over 10 million PubMed abstracts published between 1975 and 2023, we simulate the emergence of biomedical knowledge via a fine-grained daily time series. MedKGent incrementally builds the KG in a day-by-day manner using two specialized agents powered by the Qwen2.5-32B-Instruct model. The Extractor Agent identifies knowledge triples and assigns confidence scores via sampling-based estimation, which are used to filter low-confidence extractions and inform downstream processing. The Constructor Agent incrementally integrates the retained triples into a temporally evolving graph, guided by confidence scores and timestamps to reinforce recurring knowledge and resolve conflicts. The resulting KG contains 156,275 entities and 2,971,384 relational triples. Quality assessments by two SOTA LLMs and three domain experts demonstrate an accuracy approaching 90%, with strong inter-rater agreement. To evaluate downstream utility, we conduct RAG across seven medical question answering benchmarks using five leading LLMs, consistently observing significant improvements over non-augmented baselines. Case studies further demonstrate the KG's value in literature-based drug repurposing via confidence-aware causal inference.
- Open Access
- 1 July 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41 (7)
ProtGO: universal protein function prediction utilizing multi-modal gene ontology knowledge
Motivation: As one of the recalcitrant challenges in life sciences and biomedicine, protein function prediction suffers from a deluge of AI-designed proteins, particularly having to face multi-modal information in the era of big data. Importing the high-throughput neural-network-based prediction framework to replace the low-throughput biological experiments, a universal multi-modal method is straightforward in addressing the growing gap between known sequences and predicting functions. Results: To bridge the gap, we propose ProtGO, a three-step framework for predicting protein function, which leverages the credible Gene Ontology (GO) knowledge base and integrates four common modalities. Specifically, we first introduce frontier pre-trained protein language models (PLMs) for representation learning of mainstay functional protein sequences. For the remaining multi-modal data, we design a text alignment module for explainable text descriptions, a taxonomy encoding module for species-specific taxonomy, and a GO graph embedding module for biological GO relations. Each module is independent and adaptive for the referenced modalities. By harnessing these four knowledge representations, ProtGO maximizes the potential of GO resources, enhancing the performance of vanilla PLMs and biological language models (LMs) in downstream GO prediction tasks. Extensive experiments demonstrate that ProtGO significantly advances the abilities of state-of-the-art PLMs to predict protein functions: approximately 8% to 27% increase in the maximum F1 measure (Fmax) compared to base models. These comprehensive studies confirm ProtGO’s capability to deliver outstanding performance in protein function prediction by utilizing a rich blend of functional and evolutionary knowledge. Availability and implementation: Our source code and all the data are available at https://github.com/sunyatawang/ProtGO.
- 1 July 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41 (7)
Open AccessProtGO: universal protein function prediction utilizing multi-modal gene ontology knowledge
Motivation: As one of the recalcitrant challenges in life sciences and biomedicine, protein function prediction suffers from a deluge of AI-designed proteins, particularly having to face multi-modal information in the era of big data. Importing the high-throughput neural-network-based prediction framework to replace the low-throughput biological experiments, a universal multi-modal method is straightforward in addressing the growing gap between known sequences and predicting functions. Results: To bridge the gap, we propose ProtGO, a three-step framework for predicting protein function, which leverages the credible Gene Ontology (GO) knowledge base and integrates four common modalities. Specifically, we first introduce frontier pre-trained protein language models (PLMs) for representation learning of mainstay functional protein sequences. For the remaining multi-modal data, we design a text alignment module for explainable text descriptions, a taxonomy encoding module for species-specific taxonomy, and a GO graph embedding module for biological GO relations. Each module is independent and adaptive for the referenced modalities. By harnessing these four knowledge representations, ProtGO maximizes the potential of GO resources, enhancing the performance of vanilla PLMs and biological language models (LMs) in downstream GO prediction tasks. Extensive experiments demonstrate that ProtGO significantly advances the abilities of state-of-the-art PLMs to predict protein functions: approximately 8% to 27% increase in the maximum F1 measure (Fmax) compared to base models. These comprehensive studies confirm ProtGO’s capability to deliver outstanding performance in protein function prediction by utilizing a rich blend of functional and evolutionary knowledge. Availability and implementation: Our source code and all the data are available at https://github.com/sunyatawang/ProtGO.
- Top Cited Papers
- 3 April 2025
- Published by Springer Nature in Nature Methods
- Vol. 22 (5) , 1028-1039
xTrimoPGLM: unified 100-billion-parameter pretrained transformer for deciphering the language of proteins
Protein language models have shown remarkable success in learning biological information from protein sequences. However, most existing models are limited by either autoencoding or autoregressive pretraining objectives, which makes them struggle to handle protein understanding and generation tasks concurrently. We propose a unified protein language model, xTrimoPGLM, to address these two types of tasks simultaneously through an innovative pretraining framework. Our key technical contribution is an exploration of the compatibility and the potential for joint optimization of the two types of objectives, which has led to a strategy for training xTrimoPGLM at an unprecedented scale of 100 billion parameters and 1 trillion training tokens. Our extensive experiments reveal that (1) xTrimoPGLM substantially outperforms other advanced baselines in 18 protein understanding benchmarks across four categories. The model also facilitates an atomic-resolution view of protein structures, leading to an advanced three-dimensional structural prediction model that surpasses existing language model-based tools. (2) xTrimoPGLM not only can generate de novo protein sequences following the principles of natural ones, but also can perform programmable generation after supervised fine-tuning on curated sequences. These results highlight the substantial capability and versatility of xTrimoPGLM in understanding and generating protein sequences, contributing to the evolving landscape of foundation models in protein science. Trained weight for the xTrimoPGLM model, and downstream datasets are available at https://huggingface.co/biomap-research .
- 3 April 2025
- Published by Springer Nature in Nature Methods
- Vol. 22 (5) , 1028-1039
Top Cited PapersxTrimoPGLM: unified 100-billion-parameter pretrained transformer for deciphering the language of proteins
Protein language models have shown remarkable success in learning biological information from protein sequences. However, most existing models are limited by either autoencoding or autoregressive pretraining objectives, which makes them struggle to handle protein understanding and generation tasks concurrently. We propose a unified protein language model, xTrimoPGLM, to address these two types of tasks simultaneously through an innovative pretraining framework. Our key technical contribution is an exploration of the compatibility and the potential for joint optimization of the two types of objectives, which has led to a strategy for training xTrimoPGLM at an unprecedented scale of 100 billion parameters and 1 trillion training tokens. Our extensive experiments reveal that (1) xTrimoPGLM substantially outperforms other advanced baselines in 18 protein understanding benchmarks across four categories. The model also facilitates an atomic-resolution view of protein structures, leading to an advanced three-dimensional structural prediction model that surpasses existing language model-based tools. (2) xTrimoPGLM not only can generate de novo protein sequences following the principles of natural ones, but also can perform programmable generation after supervised fine-tuning on curated sequences. These results highlight the substantial capability and versatility of xTrimoPGLM in understanding and generating protein sequences, contributing to the evolving landscape of foundation models in protein science. Trained weight for the xTrimoPGLM model, and downstream datasets are available at https://huggingface.co/biomap-research .
- 4 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 44 (2) , 1058-1069
GobletNet: Wavelet-Based High-Frequency Fusion Network for Semantic Segmentation of Electron Microscopy Images
Semantic segmentation of electron microscopy (EM) images is crucial for nanoscale analysis. With the development of deep neural networks (DNNs), semantic segmentation of EM images has achieved remarkable success. However, current EM image segmentation models are usually extensions or adaptations of natural or biomedical models. They lack the full exploration and utilization of the intrinsic characteristics of EM images. Furthermore, they are often designed only for several specific segmentation objects and lack versatility. In this study, we quantitatively analyze the characteristics of EM images compared with those of natural and other biomedical images via the wavelet transform. To better utilize these characteristics, we design a high-frequency (HF) fusion network, GobletNet, which outperforms state-of-the-art models by a large margin in the semantic segmentation of EM images. We use the wavelet transform to generate HF images as extra inputs and use an extra encoding branch to extract HF information. Furthermore, we introduce a fusion-attention module (FAM) into GobletNet to facilitate better absorption and fusion of information from raw images and HF images. Extensive benchmarking on seven public EM datasets (EPFL, CREMI, SNEMI3D, UroCell, MitoEM, Nanowire and BetaSeg) demonstrates the effectiveness of our model. The code is available at https://github.com/Yanfeng-Zhou/GobletNet.
- 4 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 44 (2) , 1058-1069
GobletNet: Wavelet-Based High-Frequency Fusion Network for Semantic Segmentation of Electron Microscopy Images
Semantic segmentation of electron microscopy (EM) images is crucial for nanoscale analysis. With the development of deep neural networks (DNNs), semantic segmentation of EM images has achieved remarkable success. However, current EM image segmentation models are usually extensions or adaptations of natural or biomedical models. They lack the full exploration and utilization of the intrinsic characteristics of EM images. Furthermore, they are often designed only for several specific segmentation objects and lack versatility. In this study, we quantitatively analyze the characteristics of EM images compared with those of natural and other biomedical images via the wavelet transform. To better utilize these characteristics, we design a high-frequency (HF) fusion network, GobletNet, which outperforms state-of-the-art models by a large margin in the semantic segmentation of EM images. We use the wavelet transform to generate HF images as extra inputs and use an extra encoding branch to extract HF information. Furthermore, we introduce a fusion-attention module (FAM) into GobletNet to facilitate better absorption and fusion of information from raw images and HF images. Extensive benchmarking on seven public EM datasets (EPFL, CREMI, SNEMI3D, UroCell, MitoEM, Nanowire and BetaSeg) demonstrates the effectiveness of our model. The code is available at https://github.com/Yanfeng-Zhou/GobletNet.