Molecular property prediction stands as a cornerstone task in AI-driven drug design and discovery, wherein the atoms within a molecule serve as nodes, collectively forming a graph with bonds acting as edges. Given the crucial role of geometric structures in molecular property prediction, the integration of 3D information with various graph learning methods has been explored to enhance prediction performance. Despite the increasing adoption of the "Graph pre-training and fine-tuning" paradigm to refine molecular representations, a significant challenge persists due to the misalignment between pre-training objectives and downstream tasks. Drawing inspiration from prompt tuning techniques in Natural Language Processing (NLP), several graph prompt-based methods have emerged. However, existing approaches tend to overlook the unique properties inherent in molecular graphs. To address this gap, our paper introduces a novel approach named 3D MO lecul A rpromp T (MOAT) designed specifically for geometric molecules. Specifically, we propose atom-level prompts to capture atom distribution, geometry-level prompts tailored for molecular conformers, where different conformations have distinct chemical properties, and task-level prompts to leverage functional group properties. Results on both 3D and 2D downstream tasks demonstrate its ability to successfully bridge the data gap across diverse settings. To the best of our knowledge, this paper is the first attempt to introduce geometric graph-prompting learning for molecules.
14 October 2024 in IEEE Journal of Biomedical and Health Informatics
Gene regulatory networks (GRNs) are crucial for understanding gene regulation and cellular processes. Inferring GRNs helps uncover regulatory pathways, shedding light on the regulation and development of cellular processes. With the rise of high-throughput sequencing and advancements in computational technology, computational models have emerged as cost-effective alternatives to traditional experimental studies. Moreover, the surge in ChIPseq data for TF-DNA binding has catalyzed the development of graph neural network (GNN)-based methods, greatly advancing GRN inference capabilities. However, most existing GNN-based methods suffer from the inability to capture long-distance structural semantic correlations due to transitive interactions. In this paper, we introduce a novel GNN-based model named Hierarchical Graph Transformer with Contrastive Learning for GRN (HGTCGRN) inference. HGTCGRN excels at capturing structural semantics using a hierarchical graph Transformer, which introduces a series of gene family nodes representing gene functions as virtual nodes to interact with nodes in the GRNS. These semanticaware virtual-node embeddings are aggregated to produce node representations with varying emphasis. Additionally, we leverage gene ontology information to construct gene interaction networks for contrastive learning optimization of GRNs. Experimental results demonstrate that HGTCGRN achieves superior performance in GRN inference
Cells in the human body are regulated by sophisticated networks of gene regulation, which allows them to fulfill their cellular destiny and function. Inspired by the advancements in large language models, there have been several attempts focusing on constructing foundation models with single-cell transcriptomic data to decipher gene regulatory networks. However, these models tend to impose a sequential structure on genes within each cell, which may omit intrinsic biological characteristics and lack the utilization of other available prior knowledge. In this paper, we introduce Cell-GraphCompass (CGCompass), the pioneering foundation model that employs graph pre-training to model genes and cells. We use three types of gene-related information as node features for constructing cell graphs and collect data from three perspectives depicting relationships between genes as edge features. We pre-trained the model with over 50 million human cells and then fine-tuned it to a broad spectrum of tasks, such as batch integration, cell type annotation, single-cell gene perturbation and in silico gene knockout predictions, achieving commendable performance. Overall, CGCompass provides a practical architecture for leveraging graph pre-training to incorporate prior knowledge in constructing a foundation model for single-cell analysis.
Inspired by the advancements in pre-trained Large Language Models, there has been a surge of studies in the Life Sciences focusing on constructing foundation models with large scale single-cell RNA-seq data. These studies typically involve pre-training a transformer model on large-scale single-cell sequencing data, followed by fine-tuning for a variety of downstream tasks, achieving notable performance. However, these models all share a common short-coming: to utilize the transformer architecture, originally designed for textual data, they artificially impose a sequential structure on genes within cells, simplifying the complex interactions between genes. Furthermore, they focus solely on transcriptomic data, neglecting other relevant biological information. To address these issues, here we introduce Cell-Graph Compass (CGC), the first foundational model that leverages graph structures to model single cells and describes cells from multiple perspectives, including transcriptional profiles, gene text summaries, transcription factor regulatory networks, gene co-expression patterns, and gene positional relationships. By incorporating self-attention mechanisms, we pretrained the model on 50 million human single-cell sequencing data, resulting in a robust digital representation of cells. Extensive downstream experiments demonstrate that our approach can capture meaningful biological knowledge and achieve superior results in various problem scenarios, achieving the state-of-the-art (SOTA).
Hit-to-lead optimization must balance binding-related performance with drug-likeness, synthetic accessibility, and early safety considerations. We developed an iterative large language model agent that integrates target and starting-hit context, literature and computational tools, multi-property scoring, and feedback from previously scored analogs. A matched benchmark covered six kinase targets, five comparison strategies, a 40-candidate budget, and six independent campaigns per method. The agent achieved the highest mean aggregate score on four targets under an objective combining docking, quantitative estimate of drug-likeness (QED), synthetic accessibility, and a rule-based property panel. Re-evaluating the same candidate sets using drug-likeness-gated docking and docking-drug-likeness hypervolume produced different target-level winners, showing that metric choice changes the comparative interpretation of optimization campaigns. Candidate distributions showed higher-QED regions alongside improved docking, and ablation identified scored-candidate feedback as the most consistent contributor. In an epidermal growth factor receptor campaign, an erlotinib analog retained the 4-anilinoquinazoline core while improving docking from -6.80 to approximately -8.6 kcal mol⁻¹ and QED from 0.418 to approximately 0.73. These computational results support target-specific agent evaluation and candidate prioritization for subsequent biochemical, cellular, and ADMET characterization.