Deep learning models can accurately predict molecular properties and help making the search for potential drug candidates faster and more efficient. Many existing methods are purely data driven, focusing on exploiting the intrinsic topology and construction rules of molecules without any chemical prior information. The high data dependency makes them difficult to generalize to a wider chemical space and leads to a lack of interpretability of predictions. Here, to address this issue, we introduce a chemical element-oriented knowledge graph to summarize the basic knowledge of elements and their closely related functional groups. We further propose a method for knowledge graph-enhanced molecular contrastive learning with functional prompt (KANO), exploiting external fundamental domain knowledge in both pre-training and fine-tuning. Specifically, with element-oriented knowledge graph as a prior, we first design an element-guided graph augmentation in contrastive-based pre-training to explore microscopic atomic associations without violating molecular semantics. Then, we learn functional prompts in fine-tuning to evoke the downstream task-related knowledge acquired by the pre-trained model. Extensive experiments show that KANO outperforms state-of-the-art baselines on 14 molecular property prediction datasets and provides chemically sound explanations for its predictions. This work contributes to more efficient drug design by offering a high-quality knowledge prior, interpretable molecular representation and superior prediction performance.

Spatially resolved transcriptomics provides genetic information in space toward elucidation of the spatial architecture in intact organs and the spatially resolved cell-cell communications mediating tissue homeostasis, development, and disease. To facilitate inference of spatially resolved cell-cell communications, we here present SpaTalk, which relies on a graph network and knowledge graph to model and score the ligand-receptor-target signaling network between spatially proximal cells by dissecting cell-type composition through a non-negative linear model and spatial mapping between single-cell transcriptomic and spatially resolved transcriptomic data. The benchmarked performance of SpaTalk on public single-cell spatial transcriptomic datasets is superior to that of existing inference methods. Then we apply SpaTalk to STARmap, Slide-seq, and 10X Visium data, revealing the in-depth communicative mechanisms underlying normal and disease tissues with spatial structure. SpaTalk can uncover spatially resolved cell-cell communications for single-cell and spot-based spatially resolved transcriptomic data universally, providing valuable insights into spatial inter-cellular tissue dynamics.

Large Language Models (LLMs) have emerged as a transformative power in enhancing natural language comprehension, representing a significant stride toward artificial general intelligence. The application of LLMs extends beyond conventional linguistic boundaries, encompassing specialized linguistic systems developed within various scientific disciplines. This growing interest has led to the advent of scientific LLMs, a novel subclass specifically engineered for facilitating scientific discovery. As a burgeoning area in the community of AI for Science, scientific LLMs warrant comprehensive exploration. However, a systematic and up-to-date survey introducing them is currently lacking. In this article, we endeavor to methodically delineate the concept of “scientific language,” whilst providing a thorough review of the latest advancements in scientific LLMs. Given the expansive realm of scientific disciplines, our analysis adopts a focused lens, concentrating on the biological and chemical domains. This includes an in-depth examination of LLMs for textual knowledge, small molecules, macromolecular proteins, genomic sequences, and their combinations, analyzing them in terms of model architectures, capabilities, datasets, and evaluation. Finally, we critically examine the prevailing challenges and point out promising research directions along with the advances of LLMs. By offering a comprehensive overview of technical developments in this field, this survey aspires to be an invaluable resource for researchers navigating the intricate landscape of scientific LLMs.

Advances in single-cell RNA sequencing (scRNA-seq) have furthered the simultaneous classification of thousands of cells in a single assay based on transcriptome profiling. In most analysis protocols, single-cell type annotation relies on marker genes or RNA-seq profiles, resulting in poor extrapolation. Still, the accurate cell-type annotation for single-cell transcriptomic data remains a great challenge. Here, we introduce scDeepSort (https://github.com/ZJUFanLab/scDeepSort), a pre-trained cell-type annotation tool for single-cell transcriptomics that uses a deep learning model with a weighted graph neural network (GNN). Using human and mouse scRNA-seq data resources, we demonstrate the high performance and robustness of scDeepSort in labeling 764 741 cells involving 56 human and 32 mouse tissues. Significantly, scDeepSort outperformed other known methods in annotating 76 external test datasets, reaching an 83.79% accuracy across 265 489 cells in humans and mice. Moreover, we demonstrate the universality of scDeepSort using more challenging datasets and using references from different scRNA-seq technology. Above all, scDeepSort is the first attempt to annotate cell types of scRNA-seq data with a pre-trained GNN model, which can realize the accurate cell-type annotation without additional references, i.e. markers or RNA-seq profiles.

Large language model (LLM) based knowledge graph completion (KGC) aims to predict the missing triples in the KGs with LLMs. However, research about LLM-based KGC fails to sufficiently harness LLMs' inference proficiencies, overlooking critical structural information integral to KGs. In this paper, we explore methods to incorporate structural information into the LLMs, with the overarching goal of facilitating structure-aware reasoning. We first discuss on the existing LLM paradigms like in-context learning and instruction tuning, proposing basic structural information injection approaches. Then we propose a Knowledge Prefix Adapter (KoPA) to fulfill this stated goal. KoPA uses a structural pre-training phase to comprehend the intricate entities and relations within KGs, representing them as structural embeddings. Then KoPA communicates such cross-modal structural information understanding to the LLMs through a knowledge prefix adapter which projects the structural embeddings into the textual space and obtains virtual knowledge tokens positioned as a prefix of the input prompt. We conduct comprehensive experiments and provide incisive analysis. Our code and data are available at https://github.com/zjukg/KoPA.

Current generative knowledge graph construction approaches usually fail to capture structural knowledge by simply flattening natural language into serialized texts or a specification language. However, large generative language model trained on structured data such as code has demonstrated impressive capability in understanding natural language for structural prediction and reasoning tasks. Intuitively, we address the task of generative knowledge graph construction with code language model: given a code-format natural language input, the target is to generate triples which can be represented as code completion tasks. Specifically, we develop schema-aware prompts that effectively utilize the semantic structure within the knowledge graph. As code inherently possesses structure, such as class and function definitions, it serves as a useful model for prior semantic structural knowledge. Furthermore, we employ a rationale-enhanced generation method to boost the performance. Rationales provide intermediate steps, thereby improving knowledge extraction abilities. Experimental results indicate that the proposed approach can obtain better performance on benchmark datasets compared with baselines.¹