Pengzhi Zhang

Spatial transcriptomics links gene expression with tissue morphology, however, current tools often prioritize genomic analysis, lacking integrated image interpretation. To address this, we present Thor, a comprehensive platform for cell-level analysis of spatial transcriptomics and histological images. Thor employs an anti-shrinking Markov diffusion method to infer single-cell spatial transcriptome from spot-level data, effectively combining gene expression and cell morphology. The platform includes 10 modular tools for genomic and image-based analysis, and is paired with Mjolnir, a web-based interface for interactive exploration of gigapixel images. Thor is validated on simulated data and multiple spatial platforms (ISH, MERFISH, Xenium, Stereo-seq). Thor characterizes regenerative signatures in heart failure, screens breast cancer hallmarks, resolves fine layers in mouse olfactory bulb, and annotates fibrotic heart tissue. In high-resolution Visium HD data, it enhances spatial gene patterns aligned with histology. By bridging transcriptomic and histological analysis, Thor enables holistic tissue interpretation in spatial biology.

Artificial intelligence has revolutionized computational biology. Recent developments in omics technologies, including single-cell RNA sequencing and spatial transcriptomics, provide detailed genomic data alongside tissue histology. However, current computational models focus on either omics or image analysis, lacking their integration. To address this, we developed OmiCLIP, a visual–omics foundation model linking hematoxylin and eosin images and transcriptomics using tissue patches from Visium data. We transformed transcriptomic data into ‘sentences’ by concatenating top-expressed gene symbols from each patch. We curated a dataset of 2.2 million paired tissue images and transcriptomic data across 32 organs to train OmiCLIP integrating histology and transcriptomics. Building on OmiCLIP, our Loki platform offers five key functions: tissue alignment, annotation via bulk RNA sequencing or marker genes, cell-type decomposition, image–transcriptomics retrieval and spatial transcriptomics gene expression prediction from hematoxylin and eosin-stained images. Compared with 22 state-of-the-art models on 5 simulations, and 19 public and 4 in-house experimental datasets, Loki demonstrated consistent accuracy and robustness.

RNA velocity provides an approach for inferring cellular state transitions from single-cell RNA sequencing (scRNA-seq) data. Conventional RNA velocity models infer universal kinetics from all cells in an scRNA-seq experiment, resulting in unpredictable performance in experiments with multi-stage and/or multi-lineage transition of cell states where the assumption of the same kinetic rates for all cells no longer holds. Here we present cellDancer, a scalable deep neural network that locally infers velocity for each cell from its neighbors and then relays a series of local velocities to provide single-cell resolution inference of velocity kinetics. In the simulation benchmark, cellDancer shows robust performance in multiple kinetic regimes, high dropout ratio datasets and sparse datasets. We show that cellDancer overcomes the limitations of existing RNA velocity models in modeling erythroid maturation and hippocampus development. Moreover, cellDancer provides cell-specific predictions of transcription, splicing and degradation rates, which we identify as potential indicators of cell fate in the mouse pancreas. cellDancer enables RNA velocity estimation with cell-specific kinetics.

It is challenging to parameterize the force field for calcium ions (Ca2+) in calcium-binding proteins because of their unique coordination chemistry that involves the surrounding atoms required for stability. In this work, we observed wide variation in Ca2+ binding loop conformations of the Ca2+-binding protein calmodulin (CaM), which adopts the most populated ternary structures determined from the MD simulations, followed by ab initio quantum mechanical (QM) calculations on all twelve amino acids in the loop that coordinate Ca2+ in aqueous solution. Ca2+ charges were derived by fitting to the electrostatic potential (ESP) in the context of a classical or polarizable force field (PFF). We discovered that the atomic radius of Ca2+ in conventional force fields is too large for the QM calculation to capture the variation in the coordination geometry of Ca2+ in its ionic form, leading to unphysical charges. Specifically, we found that the fitted atomic charges of Ca2+ in the context of PFF depend on the coordinating geometry of electronegative atoms from the amino acids in the loop. Although nearby water molecules do not influence the atomic charge of Ca2+, they are crucial for compensating for the coordination of Ca2+ due to the conformational flexibility in the EF-hand loop. Our method advances the development of force fields for metal ions and protein binding sites in dynamic environments.

Protein–protein interactions drive most every biological process, but in many instances the domains mediating recognition are disordered. How specificity in binding is attained in the absence of defined structure contrasts with well-established experimental and theoretical work describing ligand binding to protein. The signaling protein calmodulin presents a unique opportunity to investigate mechanisms for target recognition given that it interacts with several hundred different targets. By advancing coarse-grained computer simulations and experimental techniques, mechanistic insights were gained in defining the pathways leading to recognition and in how target selectivity can be achieved at the molecular level. A model requiring mutually induced conformational changes in both calmodulin and target proteins was necessary and broadly informs how proteins can achieve both high affinity and high specificity.