Biology subjects

Wang, A. T.

Publications and source records attributed to Wang, A. T..

7 recordsLinked to original sources

bioRxiv2025

Pathologist-interpretable breast cancer subtyping and stratification from AI-inferred nuclear features

Artificial intelligence (AI) is making notable advances in digital pathology but faces challenges in human interpretability. Here we introduce EXPAND (EXplainable Pathologist Aligned Nuclear Discriminator), the first pathologist-interpretable AI model to predict breast cancer tumor subtypes and patient survival. EXPAND focuses on a core set of 12 nuclear pathologist-interpretable features (NPIFs), composing the Nottingham grading criteria used by the pathologists. It is a fully automated, end-to-end diagnostic workflow, which automatically extracts NPIFs given a patient tumor slide and uses them to predict tumor subtype and survival. EXPANDs performance is comparable to that of existing deep learning non-interpretable black box AI models. It achieves areas under the ROC curve (AUC) values of 0.73, 0.79 and 0.75 for predicting HER2+, HR+ and TNBC tumor subtypes, respectively, matching the performance of proprietary models that rely on substantially larger and more complex interpretable feature sets. The 12 NPIFs demonstrate strong and independent prognostic value for patient survival, underscoring their potential as biologically grounded, interpretable biomarkers for survival stratification in BC. These results lay the basis for building interpretable AI diagnostic models in other cancer indications. The complete end-to-end pipeline is made publicly available via GitHub (https://github.com/ruppinlab/EXPAND) to support community use and reproducibility.

bioinformatics↗

bioRxiv2025

Deep learning inference of cell type-specific gene expression from breast tumor histopathology

Cell type-specific gene expression from single-cell RNA sequencing (RNA-seq) is valuable for breast cancer precision oncology but available cohorts are still limited due to its high cost. Deconvolution methods infer cell type-specific expression from bulk RNA-seq at a lower cost, yet expenses and processing time of bulk RNA-seq are also non-negligible and limit their application too. To address these limitations, we developed SLIDE-EX (SLide-based Inference of DEconvolved gene EXpression), a deep-learning tool that predicts cell type-specific gene expression and abundances directly from routine breast cancer histopathology whole slide images (WSIs), using deconvolved bulk RNA-seq data as training labels. Trained on the TCGA-breast cohort and tested in cross validation and on an independent cohort of 160 cases, SLIDE-EX robustly infers the expression of thousands of genes across 9 distinct cell types, performing best for cancer associated fibroblasts and cancer cells. The abundance of these two cell types could also be robustly predicted, together with that of myeloid cells. The robustly predicted genes reflect key biological functions of their respective cell types. From a translational perspective, the inferred cell type specific expression profiles predict chemotherapy response more accurately than models based on direct prediction from the slides or from the inferred bulk expression in two independent cohorts. Going forward, SLIDE-EX is a generic approach that opens up possibilities for rapid, cost-effective cell type-specific gene expression inference in potentially any cancer type, further democratizing the characterization of the tumor microenvironment.

cancer biology↗

bioRxiv2025

Dissecting regulatory syntax in human development with scalable multiomics and deep learning

Transcription factors (TFs) establish cell identity during development by binding regulatory DNA in a sequence-specific manner, often promoting local chromatin accessibility, and regulating gene expression. Mapping accessible chromatin offers critical insights into transcriptional control, but available datasets for human development are restricted to bulk tissue, single organs, or single modalities. Here, we present the Human Development Multiomic Atlas (HDMA), a single-cell atlas of chromatin accessibility and gene expression from 817,740 fetal cells across 12 organs, spanning 203 cell types and over 1 million candidate cis-regulatory elements, many of which exhibit organ-specific in vivo enhancer activity. Deep learning models trained to predict accessibility from local DNA sequence unravel a comprehensive lexicon of motifs which influence accessibility, including composite motifs exhibiting distinct syntactic constraints predicted to mediate TF cooperativity. We identify "hard" syntactic rules requiring precise motif spacing and orientation, "soft" rules allowing flexible motif arrangements, and ubiquitous motifs that inhibit accessibility. Model-based interpretation of genetic variants reveals that disruption of motifs with positive and negative effects is associated with concordant effects on gene expression. Our work delineates how motif syntax governs cell type-specific chromatin accessibility and provides a foundational resource for decoding cis-regulatory logic and interpreting genetic variation during human development.

genomics↗

bioRxiv2023

RACER-m Leverages Structural Features for Sparse T Cell Specificity Prediction

Reliable prediction of T cell specificity against antigenic signatures is a formidable task, complicated primarily by the immense diversity of T cell receptor and antigen sequence space and the resulting limited availability of training sets for inferential models. Recent modeling efforts have demonstrated the advantage of incorporating structural information to overcome the need for extensive training sequence data, yet disentangling the heterogeneous TCR-antigen structural interface to accurately predict the MHC-allele-restricted TCR-peptide binding interactions remained challenging. Here, we present RACER-m, a coarse-grained structural template model leveraging key biophysical information from the diversity of publicly available TCR-antigen crystal structures. We find explicit inclusion of structural content substantially reduces the required number of training examples for reliable prediction of TCR-recognition specificity and sensitivity across diverse biological contexts. We demonstrate that our structural model capably identifies biophysically meaningful point-mutants that affect overall binding affinity, distinguishing its ability in predicting TCR specificity of point mutants peptides from alternative sequence-based methods. Collectively, our approach combines biophysical and inferential learning-based methods to predict TCR-peptide binding events using sparse training data. Its application is broadly applicable to studies involving both closely-related and structurally diverse TCR-peptide pairs.

biophysics↗

bioRxiv2023

Custom-engineered hydrogels for delivery of human iPSC-derived neurons into the injured cervical spinal cord

Cervical damage is the most prevalent type of spinal cord injury clinically, although few preclinical research studies focus on this anatomical region of injury. Here we present a combinatorial therapy composed of a custom-engineered, injectable hydrogel and human induced pluripotent stem cell (iPSC)-derived deep cortical neurons. The biomimetic hydrogel has a modular design that includes a protein-engineered component to allow customization of the cell-adhesive peptide sequence and a synthetic polymer component to allow customization of the gel mechanical properties. In vitro studies with encapsulated iPSC-neurons were used to select a bespoke hydrogel formulation that maintains cell viability and promotes neurite extension. Following injection into the injured cervical spinal cord in a rat contusion model, the hydrogel biodegraded over six weeks without causing any adverse reaction. Compared to cell delivery using saline, the hydrogel significantly improved the reproducibility of cell transplantation and integration into the host tissue. Across three metrics of animal behavior, this combinatorial therapy significantly improved sensorimotor function by six weeks post transplantation. Taken together, these findings demonstrate that design of a combinatorial therapy that includes a gel customized for a specific fate-restricted cell type can induce regeneration in the injured cervical spinal cord.

bioengineering↗

bioRxiv2019

Distinct roles of BRCA2 in replication fork protection in response to hydroxyurea and DNA interstrand crosslinks

DNA interstrand crosslinks (ICLs) are a form of DNA damage that requires the interplay of a number of repair proteins including those of the Fanconi anemia (FA) and the homologous recombination (HR) pathways. Pathogenic variants in the essential gene BRCA2/FANCD1, when monoallelic, predispose to breast and ovarian cancer, and when biallelic, results in a severe subtype of Fanconi anemia. BRCA2 function in the FA pathway is attributed to its role as a mediator of the RAD51 recombinase in HR repair of the programmed DNA double strand breaks (DSB). BRCA2 and RAD51 functions are also required to protect stalled replication forks from nucleolytic degradation during response to hydroxyurea (HU). While RAD51 has been shown to be necessary in the early steps of ICL repair to prevent aberrant nuclease resection, the role of BRCA2 in this process has not been described. Here, based on the analysis of BRCA2 DNA binding domain (DBD) mutants discovered in FA patients presenting with atypical FA-like phenotypes, we establish that BRCA2 is necessary for protection of DNA at an ICL. Cells carrying DBD BRCA2 mutations are sensitive to ICL inducing agents but resistant to HU treatment consistent with relatively high HR repair in these cells. BRCA2 function at an ICL protects against DNA2-WRN nuclease-helicase complex and not the MRE11 nuclease implicated in the resection of HU-stalled replication forks. Our results also indicate that unlike the processing at HU-stalled forks, function of the SNF2 translocases (SMARCAL1, ZRANB3, or HLTF), implicated in fork reversal, are not an integral component of the ICL repair, pointing to a different mechanism of fork protection at different DNA lesions.

molecular biology↗

bioRxiv2019

Allele-Specific QTL Fine-Mapping with PLASMA

Although quantitative trait locus (QTL) associations have been identified for many molecular traits such as gene expression, it remains challenging to distinguish the causal nucleotide from nearby variants. In addition to traditional QTLs by association, allele-specific (AS) QTLs are a powerful measure of cis-regulation that are largely concordant with traditional QTLs, and can be less susceptible to technical/environmental noise. However, existing asQTL analysis methods do not produce probabilities of causality for each marker, and do not take into account correlations among markers at a locus in linkage disequilibrium (LD). We introduce PLASMA (PopuLation Allele-Specific MApping), a novel, LD-aware method that integrates QTL and asQTL information to fine-map causal regulatory variants while drawing power from both the number of individuals and the number of allelic reads per individual. We demonstrate through simulations that PLASMA successfully detects causal variants over a wide range of genetic architectures. We apply PLASMA to RNA-Seq data from 524 kidney tumor samples and show that over 17 percent of loci can be fine-mapped to within 5 causal variants, compared less than 2 percent of loci using existing QTL-based fine-mapping. PLASMA furthermore achieves a greater power at 50 samples than conventional QTL fine-mapping does at over 500 samples. Overall, PLASMA achieves a 6.9-fold reduction in median 95% credible set size compared to existing QTL-based fine-mapping. We additionally apply PLASMA to H3K27AC ChIP-Seq from 28 prostate tumor/normal samples and demonstrate that PLASMA is able to prioritize markers even at small samples, with PLASMA achieving a 1.3-fold reduction in median 95% credible set sizes over existing QTL-based fine-mapping. Variants in the PLASMA credible sets for RNA-Seq and ChIP-Seq were enriched for open chromatin and chromatin looping (respectively) at a comparable or greater degree than credible variants from existing methods, while containing far fewer markers. Our results demonstrate how integrating AS activity can substantially improve the detection of causal variants from existing molecular data and at low sample size.

bioinformatics↗