Yibo Peng, James Song, Lei Li, Xinyu Yang, Mihai Christodorescu, Ravi Mangal, Corina S. Pasareanu, Haizhong Zheng, Beidi Chen. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Spatial transcriptomics enables the study of gene expression within the spatial context of tissues, offering valuable insights into tissue organization and function. However, technical limitations can result in large missing regions of data, which hinder accurate downstream analyses and biological interpretation. To address these challenges, we propose DISCO (DIffusion model for Spatial transcriptomics data COmpletion), a framework with three key features. First, DISCO employs a graph neural network-based region encoder to integrate spatial and gene expression information from observed regions, generating latent representations that guide the prediction of missing regions. Second, it uses two diffusion modules: a position diffusion module to predict the spatial layout of missing regions, and a gene expression diffusion module to generate gene expression profiles conditioned on the predicted coordinates. Third, DISCO incorporates neighboring region information during inference to guide the denoising process, ensuring smooth transitions and biologically coherent results. We validate DISCO across multiple sequencing platforms, species, and datasets, demonstrating its effectiveness in reconstructing large missing regions. DISCO is implemented as open-source software, providing researchers with a powerful tool to enhance data completeness and advance spatial transcriptomics research.

Explainable AI (XAI) has gained significant attention for providing insights into the decision-making processes of deep learning models, particularly for image classification tasks through saliency-based visual explanations. Despite their success, key challenges persist due to the scarcity of annotated datasets and the absence of standardized evaluation protocols. In this paper, we introduce Saliency-Bench, a novel benchmark designed to evaluate visual explanations generated by saliency methods across multiple datasets. We curated, constructed, and annotated eight datasets, each covering diverse tasks such as scene classification, cancer diagnosis, object classification, and action classification, with corresponding ground-truth explanation annotations. The benchmark includes a standardized and unified evaluation pipeline for assessing faithfulness and alignment of the visual explanation, providing a holistic visual explanation performance assessment. We benchmark these eight datasets with widely used saliency methods on different image classifier architectures to evaluate explanation quality. Additionally, we developed an user-friendly toolkit for automating the evaluation pipeline, from data accessing, and data loading, to result evaluation. The benchmark is available at https://github.com/XAIdataset/XAIdataset.github.io.

The Immunoglobulin fold (Ig-fold) is found in proteins from all domains of life and represents the most populous fold in the human genome, with current estimates ranging from 2 to 3% of protein coding regions. That proportion is much higher in the surfaceome where Ig and Ig-like domains orchestrate cell-cell recognition, adhesion and signaling. The ability of Ig-domains to reliably fold and self-assemble through highly specific interfaces represents a remarkable property of these domains, making them key elements of molecular interaction systems: the immune system, the nervous system, the vascular system and the muscular system. We define a universal residue numbering scheme, common to all domains sharing the Ig-fold in order to study the wide spectrum of Ig-domain variants constituting the Ig-proteome and Ig-Ig interactomes at the heart of these systems. The "IgStrand numbering scheme" enables the identification of Ig structural proteomes and interactomes in and between any species, and comparative structural, functional, and evolutionary analyses. We review how Ig-domains are classified today as topological and structural variants and highlight the "Ig-fold irreducible structural signature" shared by all of them. The IgStrand numbering scheme lays the foundation for the systematic annotation of structural proteomes by detecting and accurately labeling Ig-, Ig-like and Ig-extended domains in proteins, which are poorly annotated in current databases and opens the door to accurate machine learning. Importantly, it sheds light on the robust Ig protein folding algorithm used by nature to form beta sandwich supersecondary structures. The numbering scheme powers an algorithm implemented in the interactive structural analysis software iCn3D to systematically recognize Ig-domains, annotate them and perform detailed analyses comparing any domain sharing the Ig-fold in sequence, topology and structure, regardless of their diverse topologies or origin. The scheme provides a robust fold detection and labeling mechanism that reveals unsuspected structural homologies among protein structures beyond currently identified Ig- and Ig-like domain variants. Indeed, multiple folds classified independently contain a common structural signature, in particular jelly-rolls. Examples of folds that harbor an "Ig-extended" architecture are given. Applications in protein engineering around the Ig-architecture are straightforward based on the universal numbering.

Influence maximization (IM) is formulated as selecting a set of initial users from a social network to maximize the expected number of influenced users. Researchers have made great progresses to design various traditional methods, yet both theoretical design and performance gain are close to their limits. In the past few years, learning-based IM methods have emerged to achieve stronger generalization ability to unknown graphs than traditional ones. However, the development of learning-based IM methods is still limited by fundamental obstacles, including 1) the difficulty of effectively solving the objective function; 2) the difficulty of characterizing the diversified and underlying diffusion patterns; and 3) the difficulty of adapting the solution under various node-centrality-constrained IM variants. To cope with the above challenges, we design a novel framework DeepIM to generatively characterize the latent representation of seed sets, and we propose to learn the diversified information diffusion pattern in a data-driven and end-to-end manner. Finally, we design a novel objective function to infer optimal seed sets under flexible node-centrality-based budget constraints. Extensive analyses are conducted over both synthetic and real-world datasets to demonstrate the overall performance of DeepIM.