Abstract Motivation Large language models (LLMs) have demonstrated strong general reasoning abilities, but applying them to domain-specific tasks such as analysing single-cell RNA sequencing data remains a challenge. A central task in this domain is cell type annotation, which is critical for understanding cellular heterogeneity. Although recent foundation models attempt to automate this process, they typically annotate cells independently, without considering batch-level context or providing explanatory reasoning. To address this limitation, we introduce the CellPuzzles benchmark, which reformulates cell type annotation as a batch-level reasoning task. CellPuzzles spans diverse tissues, diseases, and donor conditions, and requires reasoning across the batch-level cellular context to ensure label uniqueness. Results We find that off-the-shelf LLMs struggle on this task, with the best baseline (OpenAI o1) achieving only 19.0% batch-level accuracy. To fill this gap, we propose Cell-o1, a 7B LLM trained via supervised fine-tuning on distilled reasoning traces, followed by reinforcement learning with batch-level rewards. Cell-o1 achieves state-of-the-art performance, outperforming OpenAI o1 by over 73% and generalizing well across contexts. Further analysis of training dynamics and reasoning behaviors provides insights into batch-level annotation performance and emergent expert-like reasoning. Availability and Implementation Code and data are available at https://github.com/ncbi-nlp/cell-o1.

This paper proposes joint sensing and beamforming solutions tailored for extremely large-scale MIMO (XL-MIMO) near-field systems under both static and dynamic scenarios. For static scenarios, we develop a novel Multi-Layer Reconstruction (MLR) mechanism to address the challenges of large-scale near-field dictionary matrix and coarse range grid spacing, and further propose a sparse sensing algorithm, MLR mechanism based Linear Approximation Variational Bayesian Inference (MLR-LA-VBI), to achieve precise user/target position and radar cross section (RCS) sensing with low pilot overhead. Building upon these sensing results, a beamforming scheme is proposed to optimize radiation patterns. For dynamic scenarios, we exploit near-field Doppler-frequency characteristics to propose the modified MLR-LA-VBI (MMLR-LA-VBI) algorithm for sensing and a predictive beamforming framework, where the former serves as the core module of the latter. Our sensing approach enables full motion status sensing of users/targets from a single echo without requiring prior information of the target motion model. By eliminating echo accumulation and leveraging correlations across consecutive coherent processing intervals (CPIs), it achieves high performance with low computational complexity. Moreover, the proposed predictive beamforming framework naturally inherits the aforementioned advantages of MMLR-LA-VBI, and leverages the sensed full motion status to achieve an efficient and seamless beam tracking scheme with Doppler frequency compensation. In addition, theoretical analysis is conducted to characterize the algorithmic complexity, highlighting the advantages of proposed algorithms in terms of efficiency. Finally, simulations and analyses validate the effectiveness of the proposed algorithms in both static and dynamic scenarios.

Reconfigurable intelligent surface (RIS) plays an essential role in alleviating severe path loss in millimeter wave communication systems. Its performance hinges on the precise modeling of high-dimensional cascaded channels. However, traditional modeling approaches require extensive experience in radio propagation, resulting in complex and inefficient processes. To overcome these limitations, we transform the RIS channel modeling into a channel distribution transport mapping problem and introduce a generative model based on rectified flow. Our approach integrates distance information into a diffusion transformer (DiT) architecture through cross-attention mechanisms, resulting in a conditional DiT capable of synthesizing target channels from distance inputs. We further optimize the rectified flow into a single-step generator via reflow techniques. Building on this framework, we design a generative digital twin (DT) channel model that serves as a high fidelity data generator for downstream tasks. The proposed model acts as a virtual replica of the propagation environment, enabling efficient channel data synthesis for training communication algorithms such as channel state information feedback and channel estimation. Simulation results show that our approach generates channels with minimal distribution discrepancy compared to real channels (a maximum mean discrepancy < 0.01), outperforming existing generative methods. Furthermore, the reflow-driven DT channel model achieves the shortest generation time among all evaluated benchmarks.

Extremely large-scale MIMO (XL-MIMO) is regarded as a pivotal enabler for achieving ultra-high spectral efficiency in 6G communications. Near-field channel models, which integrate both line-of-sight (LoS) and non-line-of-sight (NLoS) components, provide accurate characterizations of near-field XL-MIMO channels. However, existing channel estimation schemes encounter severe performance bottlenecks due to the high-dimensional nature of near-field XL-MIMO channels and their structured angular sparsity compared to far-field MIMO systems. To address these challenges, we propose a variance exploding stochastic differential equation (VE-SDE) generator based on an improved diffusion transformer (IDiT) network. The VE-SDE progressively maps the complex XL-MIMO channel distribution to a tractable prior distribution by gradually injecting noise. We utilize the patchify technique to decompose the perturbed angular domain channels into token sequences, which are then processed with diffusion transformer (DiT) blocks, substantially reducing floating-point operations (FLOPs). Additionally, a sparse self-attention mechanism is employed to enhance structured sparsity characterization learning, thereby improving estimation accuracy. Theoretical analysis and numerical experiments show that the VE-SDE generator exhibits strong generalizability and robustness across diverse channel distributions without requiring retraining. Simulation results reveal that the proposed method outperforms state-of-the-art estimation approaches, achieving high-fidelity channel estimation with only 20% pilot density.

The rise of Multi-modal Pre-training highlights the necessity for a unified Multi-Modal Knowledge Graph (MMKG) representation learning framework. Such a framework is essential for embedding structured knowledge into multi-modal Large Language Models effectively, alleviating issues like knowledge misconceptions and multi-modal hallucinations. In this work, we explore the efficacy of models in accurately embedding entities within MMKGs through two pivotal tasks: Multi-modal Knowledge Graph Completion (MKGC) and Multi-modal Entity Alignment (MMEA). Building on this foundation, we propose a novel SNAG method that utilizes a Transformer-based architecture equipped with modality-level noise masking to robustly integrate multi-modal entity features in KGs. By incorporating specific training objectives for both MKGC and MMEA, our approach achieves SOTA performance across a total of ten datasets, demonstrating its versatility. Moreover, SNAG can not only function as a standalone model but also enhance other existing methods, providing stable performance improvements. Code and data are available at https://github.com/zjukg/SNAG.

The emergence of extremely large-scale MIMO (XLMIMO) has made target sensing in near-field environments crucial. However, the vast number of antennas and the big size of near-field dictionary matrix (DM) result in substantial pilot overhead for beam training algorithms and significant computational complexity for subspace algorithms. Moreover, there is few of work capable of accurately obtaining information beyond target location, such as radar cross-section (RCS). To this end, we propose a high-precision, low-pilot-overhead off-grid compressive sensing (CS) algorithm capable of jointly estimating target's location and RCS-the Multi-Layer Reconstruction Linear Approximation Variational Bayesian Inference (MLR-LA-VBI) algorithm. Specifically, the entire algorithm is divided into two phases. In the first phase, we propose the Multi-Layer Reconstruction (MLR) mechanism to reconstruct a surrogate DM. In the second phase, based on the surrogate DM, thus proposing the MLR-LA-VBI algorithm for joint estimation of target location and RCS. The final simulation results verify the superior performance of the proposed algorithm.

The generation of molecules with desired properties has become increasingly popular, revolutionizing the way scientists design molecular structures and providing valuable support for chemical and drug design. However, despite the potential of language models in molecule generation, they face challenges such as generating syntactically or chemically flawed molecules, having narrow domain focus, and struggling to create diverse and feasible molecules due to limited annotated data or external molecular databases. To tackle these challenges, we introduce MolGen, a pre-trained molecular language model tailored specifically for molecule generation. Through the reconstruction of over 100 million molecular SELFIES, MolGen internalizes structural and grammatical insights. This is further enhanced by domain-agnostic molecular prefix tuning, fostering robust knowledge transfer across diverse domains. Importantly, our chemical feedback paradigm steers the model away from "molecular hallucinations", ensuring alignment between the model's estimated probabilities and real-world chemical preferences. Extensive experiments on well-known benchmarks underscore MolGen's optimization capabilities in properties such as penalized logP, QED, and molecular docking. Additional analyses confirm its proficiency in accurately capturing molecule distributions, discerning intricate structural patterns, and efficiently exploring the chemical space (https://github.com/zjunlp/MolGen).

Large Language Models (LLMs), with their remarkable task-handling capabilities and innovative outputs, have catalyzed significant advancements across a spectrum of fields. However, their proficiency within specialized domains such as biomolecular studies remains limited. To address this challenge, we introduce Mol-Instructions, a comprehensive instruction dataset designed for the biomolecular domain. Mol-Instructions encompasses three key components: molecule-oriented instructions, protein-oriented instructions, and biomolecular text instructions. Each component aims to improve the understanding and prediction capabilities of LLMs concerning biomolecular features and behaviors. Through extensive instruction tuning experiments on LLMs, we demonstrate the effectiveness of Mol-Instructions in enhancing large models' performance in the intricate realm of biomolecular studies, thus fostering progress in the biomolecular research community. Mol-Instructions is publicly available for ongoing research and will undergo regular updates to enhance its applicability (https://github.com/zjunlp/Mol-Instructions).

Recent embedding-based methods have achieved great successes in exploiting entity alignment from knowledge graph (KG) embeddings of multiple modalities. In this paper, we study embedding-based entity alignment (EEA) from a perspective of generative models. We show that EEA shares similarities with typical generative models and prove the effectiveness of the recently developed generative adversarial network (GAN)-based EEA methods theoretically. We then reveal that their incomplete objective limits the capacity on both entity alignment and entity synthesis (i.e., generating new entities). We mitigate this problem by introducing a generative EEA (GEEA) framework with the proposed mutual variational autoencoder (M-VAE) as the generative model. M-VAE enables entity conversion between KGs and generation of new entities from random noise vectors. We demonstrate the power of GEEA with theoretical analysis and empirical experiments on both entity alignment and entity synthesis tasks. The source code and datasets are available at github.com/zjukg/GEEA.

Zero-shot learning (ZSL) aims to predict unseen classes whose samples have never appeared during training. One of the most effective and widely used semantic information for zero-shot image classification are attributes which are annotations for class-level visual characteristics. However, the current methods often fail to discriminate those subtle visual distinctions between images due to not only the shortage of fine-grained annotations, but also the attribute imbalance and co-occurrence. In this paper, we present a transformer-based end-to-end ZSL method named DUET, which integrates latent semantic knowledge from the pre-trained language models (PLMs) via a self-supervised multi-modal learning paradigm. Specifically, we (1) developed a cross-modal semantic grounding network to investigate the model's capability of disentangling semantic attributes from the images; (2) applied an attribute-level contrastive learning strategy to further enhance the model's discrimination on fine-grained visual characteristics against the attribute co-occurrence and imbalance; (3) proposed a multi-task learning policy for considering multi-model objectives. We find that our DUET can achieve state-of-the-art performance on three standard ZSL benchmarks and a knowledge graph equipped ZSL benchmark. Its components are effective and its predictions are interpretable.

Molecular property is usually observed with a limited number of samples, and researchers have considered property prediction as a few-shot problem. One important fact that has been ignored by prior works is that each molecule can be recorded with several different properties simultaneously. To effectively utilize many-to-many correlations of molecules and properties, we propose a Graph Sampling-based Meta-learning (GS-Meta) framework for few-shot molecular property prediction. First, we construct a Molecule-Property relation Graph (MPG): molecule and properties are nodes, while property labels decide edges. Then, to utilize the topological information of MPG, we reformulate an episode in meta-learning as a subgraph of the MPG, containing a target property node, molecule nodes, and auxiliary property nodes. Third, as episodes in the form of subgraphs are no longer independent of each other, we propose to schedule the subgraph sampling process with a contrastive loss function, which considers the consistency and discrimination of subgraphs. Extensive experiments on 5 commonly-used benchmarks show GS-Meta consistently outperforms state-of-the-art methods by 5.71%-6.93% in ROC-AUC and verify the effectiveness of each proposed module. Our code is available at https://github.com/HICAI-ZJU/GS-Meta.

Multi-modal entity alignment (MMEA) aims to discover identical entities across different knowledge graphs (KGs) whose entities are associated with relevant images. However, current MMEA algorithms rely on KG-level modality fusion strategies for multi-modal entity representation, which ignores the variations of modality preferences of different entities, thus compromising robustness against noise in modalities such as blurry images and relations. This paper introduces MEAformer, a mlti-modal entity alignment transformer approach for meta modality hybrid, which dynamically predicts the mutual correlation coefficients among modalities for more fine-grained entity-level modality fusion and alignment. Experimental results demonstrate that our model not only achieves SOTA performance in multiple training scenarios, including supervised, unsupervised, iterative, and low-resource settings, but also has a limited number of parameters, efficient runtime, and interpretability. Our code is available at https://github.com/zjukg/MEAformer.

Molecular representation learning contributes to multiple downstream tasks such as molecular property prediction and drug design. To properly represent molecules, graph contrastive learning is a promising paradigm as it utilizes self-supervision signals and has no requirements for human annotations. However, prior works fail to incorporate fundamental domain knowledge into graph semantics and thus ignore the correlations between atoms that have common attributes but are not directly connected by bonds. To address these issues, we construct a Chemical Element Knowledge Graph (KG) to summarize microscopic associations between elements and propose a novel Knowledge-enhanced Contrastive Learning (KCL) framework for molecular representation learning. KCL framework consists of three modules. The first module, knowledge-guided graph augmentation, augments the original molecular graph based on the Chemical Element KG. The second module, knowledge-aware graph representation, extracts molecular representations with a common graph encoder for the original molecular graph and a Knowledge-aware Message Passing Neural Network (KMPNN) to encode complex information in the augmented molecular graph. The final module is a contrastive objective, where we maximize agreement between these two views of molecular graphs. Extensive experiments demonstrated that KCL obtained superior performances against state-of-the-art baselines on eight molecular datasets. Visualization experiments properly interpret what KCL has learned from atoms and attributes in the augmented molecular graphs.

In the fact of growing number of cases, Chinese courts have gradually formed a trial mode to improve the efficiency of trials by conducting trials around the controversial issues. However, identifying the controversy issue in specific cases is not only affected by the uncertainty of facts and laws, but also by the discretion of the judges and extra-case factors, and cannot be expressed as a standard format, which lead to the controversial issues based case retrieval a challenge problem. In this paper, we propose a controversial issues merging algorithm based on K-means clustering for Chinese legal texts. The proposed algorithm can determine the number of clusters of the given cause of action automatically and merge the controversial issues semantically, which makes the case information retrieval more accurate and effective.