Biology subjects

Niazi, M. K. K.

Publications and source records attributed to Niazi, M. K. K..

11 recordsLinked to original sources

bioRxiv2026

OmniT2I: a unified table-to-image framework for survival prediction and cross-modal inference with incomplete multimodal cancer data

Molecular profiles and histopathology provide complementary information about cancer prognosis, but these data are often incomplete for individual patients. Here we present OmniT2I, a table-to-image framework for survival prediction and cross-modal inference from incomplete multimodal cancer data. Each molecular profile was converted into a three-channel image representation with correlation-based feature placement. OmniT2I combines available molecular and histopathology embeddings through availability-aware fusion and predicts missing modalities from those observed. Across 10,373 patients with 32 cancer types, the C-index was 0.631 within cancer types and 0.783 with all cancers pooled (0.660 and 0.787 in 5,605 patients with all five modalities). A single model handled all 31 modality combinations and, compared with nine survival methods trained separately for each, achieved higher C-indices in 240 of 279 pooled and 226 of 279 within-cancer comparisons. Adding predictions of withheld modalities to observed inputs improved pooled C-indices in all 180 pairings of observed and withheld modality sets and within-cancer C-indices in 164. Adding measured histopathology to RNA produced the largest within-cancer C-index gain among the four additional modalities evaluated, whereas adding copy number variation reduced it. In seven external cohorts, frozen RNA-only C-indices differed from test performance in the corresponding cancer type in TCGA by a median of -0.003 (range, -0.092 to +0.079). OmniT2I enables prognostic modeling with incomplete multimodal data.

bioinformatics↗

bioRxiv2026

Weak supervision of H&E slides reveals systems-level biology and functional states that govern therapeutic resistance

Precision oncology lacks scalable methods to identify the mechanisms that mediate therapeutic resistance for individual patients. Resistance often arises from focal cellular niches that are obscured by bulk profiling and costly to resolve with multi-omics. Here, we show that deep learning (DL), applied to routine histology, can localize focal tissue regions enriched for therapeutically relevant biology. Using 3111 breast cancer H&E slides with matched bulk transcriptomics, we trained weakly-supervised DL models to infer activities of immune, metabolic, and tumor-intrinsic phenotypes implicated in therapeutic resistance (AUROC>0.80; PCC>0.64). Accurate inference of these phenotypes should identify tissue regions enriched for the corresponding biological signal. Therefore, we validated phenotype inference and spatial localization with complementary analyses. Tissue-matched multiplexed immunofluorescence showed concordance between inferred immune states and corresponding cell fractions (p=0.006-0.106). Across multi-institutional cohorts, model-derived phenotypes recovered expected relationships with therapeutic outcomes (p<0.045). Finally, in a blinded evaluation, pathologists confirmed that model-derived high-attention regions were enriched for phenotype-specific morphology (p<2.408*10-5). Because evaluated phenotypes represent diverse mechanisms of resistance across therapeutic modalities, these findings provide a foundation for resistance-directed localization using therapeutic outcomes as supervision. By directing deep profiling toward model-prioritized regions, this framework could enable scalable nomination of candidate mediators of resistance for subsequent functional validation across real-world patient populations. One sentence summaryWeakly supervised deep learning localizes focal tissue regions enriched for therapeutically relevant biology in routine histology, thus offering a scalable strategy to study therapeutic resistance across large patient populations.

biophysics↗

bioRxiv2026

Personalized Morphology, Replication Timing, and RNA based Gene Expression Networks for Basal-like and Classical subtyping genes in Pancreatic Adenocarcinoma

Network biology traditionally identifies gene correlations that reflect biological pathways. While LIONESS enables individualized gene networks, the influence of replication timing on these correlations remains unexplored. Replication timing reflects the temporal order of DNA synthesis and is tightly linked to chromatin state, methylation, and transcriptional stability, all of which affect tumor behavior. Integrating replication-timing proxies derived from methylation data therefore offers a bridge between epigenetic state and functional gene coordination, while morphology provides an additional route for inferring gene expression. This is the first study to integrate replication-timing proxies and morphological embeddings into individualized LIONESS gene networks. The aim is to determine how replication timing and morphology derived from bulk methylation and image embeddings influence gene coexpression in pancreatic cancer. Patient-specific networks were generated for basal and classical pancreatic ductal adenocarcinoma subtypes using TCGA data. Results show an 80% AUC for RNA-replication-timing-based subtype prediction modules and a 75% AUC for morphology-based networks. Incorporating replication timing and morphology increased network robustness while maintaining classification performance. Notably, the 80% AUC was achieved using only 17 of the 50 Moffitt genes, with 16 overlapping the PURIST gene set, indicating that replication timing captures clinically relevant regulatory structure. These findings suggest that replication-timing proxies can act as epigenetic indicators of mechanistic gene coordination and may help identify patients with distinct replication stress or chromatin accessibility profiles relevant to therapeutic response.

bioinformatics↗

bioRxiv2026

Neural Networks as Entropic Systems: Applications in Digital Pathology

Deep learning systems in digital pathology are widely regarded as opaque, limiting clinical trust and interpretability. We present a framework for empirically characterizing training-time learning dynamics in neural networks by directly measuring activation structure, weight evolution, and spectral organization during optimization. Using TCGA-BRCA whole-slide images with replication-timing-derived methylation proxies as regression targets, we trained a Vision Transformer and tracked its intra-epoch behavior across 20 epochs. We observed reproducible structural signatures during training. Correlated groups of neurons formed stable activation modules whose modularity increased as training progressed, accompanied by a reduction in representation entropy of up to 60%. Weight trajectories exhibited bounded diffusion with progressively reduced variance, consistent with a damped stochastic process, and converged toward a stable stationary regime in later epochs. In image space, model attention systematically shifted from collagen-rich stromal regions in early epochs to basophilic, proliferative nuclear regions in later epochs, aligning with known histologic correlates of replicative stress. These findings demonstrate that neural networks develop predictable, quantifiable internal structure during training that can be directly visualized and measured. Framing learning dynamics in terms of entropy, modular organization, and stochastic stabilization provides a practical, mechanistic lens for interpreting how pathology AI models acquire biologically meaningful representations.

pathology↗

bioRxiv2026

Performance of Naiive Spectral Geometric Models in Histopathology AI

There have been no systematic evaluations of purely spectral models for digital pathology tasks. We implemented and benchmarked four pipelines: binary classification on the BreaKHis dataset, multi-class region classification in glioblastoma, spatial transcriptomics, and denoising on Visium 10x. Across all tasks, extensive cross-validation and grouped splits showed that purely spectral models did not improve performance over CNN-only baselines, but offer useful complementary tools for interpretability and processing. Denoising showed strong performance that proves utility in data-scarce or heterogeneous image environments. Equivalence testing confirms that spectral and CNN model performances fall outside {+/-}3% AUC. Fusion models between CNNs and spectral models show higher balanced accuracy. Spectral models failed to generalize across spatial transcriptomics tasks, with low correlation despite stable training loss. These findings represent a systematic negative result: despite their theoretical richness, spectral geometric features and SNO embeddings prove to be complementary features for WSI classification or segmentation. Reporting such outcomes is essential to establish empirical boundaries for spectral methods and to encourage future work on conditions or data modalities where these approaches may hold greater promise.

pathology↗

bioRxiv2026

Deep Learning-Based Spatial Immunoprofiling of Multiplex Immunofluorescence Images Distinguishes Tuberculosis Disease States in Diversity Outbred Mice

Tuberculosis (TB), caused by Mycobacterium tuberculosis (M.tb), remains a major global health challenge, with approximately 10.8 million new cases and 1.25 million deaths reported in 2023. Human responses to M.tb are heterogeneous with clinical outcomes including bacterial clearance, asymptomatic latent M.tb infection, and severe fatal pulmonary TB. Here, we aim to address knowledge gaps in the organization of M.tb granulomas by identifying cell-based spatial features indicating asymptomatic lung infection. To address this gap, which cannot be directly studied in humans, we used lung sections from M.tb infected Diversity Outbred mice with acute pulmonary TB, asymptomatic M.tb infection, or chronic pulmonary TB that were stained for T cells, B cells, macrophages, and bronchiolar epithelial cells by multiplexed immunofluorescence. We first developed a new, accurate model to automatically segment lung granulomas, detect/quantify the cell types within granulomas, and extract the location of immune cells in granulomas for each disease state. Analysis of model-derived results show that lung granulomas from asymptomatic mice have a characteristic spatial profile consisting of higher CD4+ and CD8a+ T cell densities, closer B cell proximity to bronchiolar epithelium, and increased T cell-macrophage proximity in asymptomatic M.tb infection. Next, we propose a second novel approach to utilize a large language model (LLM) to independently decode complex cellular patterns within granulomas, and distinguished key immunological signatures: balanced immune expression in asymptomatic M.tb mice, dysfunctional responses with low cellularity in acute TB, and highest immune cells in chronic TB. Overall, the results from this study show that lung granulomas in asymptomatic infection are characterized by increased T cell density, increased numbers of peribronchiolar B cells, and T cells closer to macrophages as compared to acute and chronic TB. These methods and results help establish an automated pipeline to extract and analyze data from multiplexed fluorescence images and provide a foundation to better understand how granuloma architecture varies by disease state.

immunology↗

bioRxiv2026

Spatial Regression of Morphology-Protein Coupling in Tumour Proteomics

Spatial proteomics has enabled high-resolution characterization of protein organization within tumor microenvironments, yet most computational approaches implicitly assume spatial homogeneity and focus on clustering rather than diffusion constraints imposed by tissue morphology. Here, we model morphology-protein coupling in triplenegative breast cancer using geographically weighted regression (GWR) applied to 41 publicly available Multiplexed Ion Beam Imaging (MIBI) samples comprising 36 protein markers. Single-cell morphometric features were extracted from MIBI spots and combined with spatial adjacency graphs to model location-specific protein dispersion. Compared with ordinary least squares and ridge regression baselines, GWR consistently demonstrated superior performance across regression metrics, explaining substantially greater spatial variance in protein intensity (+.4 R2 improvements across markers) while reducing mean absolute and squared errors. Information-theoretic analysis showed lower (Aikake Information Criterion Corrected) AICc values for GWR across the majority of markers, indicating improved model fit. Spatial autocorrelation diagnostics further confirmed that GWR residuals exhibited near-random structure, with significant reductions in Morans I and Gearys C relative to global models, demonstrating effective capture of local heterogeneity. Eight proteins with significant spatial autocorrelation, including B7-H3 and -catenin, showed pronounced morphology-dependent dispersion patterns that were not recoverable using global regression. These results demonstrate that explicitly modeling spatial heterogeneity yields more accurate and interpretable representations of protein organization and supports a diffusion-barrier view of pathoproteomics beyond agglomeration alone.

bioinformatics↗

bioRxiv2025

Quantum Chemical Profiling of Protein Mutations via Fragment-Based DFT

Missense mutations have been extensively studied in tumor-suppressing antigens (TP53) to understand oncogenesis within malignant epithelial cells. Using Whole Exome Sequencing (WXS), missense mutations can be profiled into protein sequences to identify the most common variants in tumor samples. Since nearly 80% of mutations arise randomly, it is necessary to isolate those that produce dysfunctional proteins within large cohorts. Using threading and generative algorithms such as AlphaFold and ColabFold, large cohorts of WXS information can be converted into computationally analyzable structures. By evaluating both highand low-confidence regions in these structures, these antigens can be studied en masse using pipelines that generate analytical inputs for quantum chemistry analysis. we created a pipeline that processed whole-exome sequencing (WXS) data and selected 28 representative TP53 missense mutants from the TCGA-BRCA cohort for quantum-chemical feasibility analysis. These structures were systematically cleaned using tools such as OpenBabel and AmberTools, and each was prepared for Natural Population Analysis (NPA), Electrostatic Potential (ESP) calculations, and Highest and Lowest Occupied Molecular Orbital (HOMO/LUMO) evaluation within Q-Chem. Using this pipeline, population genomics can be integrated with chemoinformatics to analyze electron density concentrations and generate hypothesis-generating electronic descriptors associated with protein dysfunction. By modifying the generated inputs, additional analyses such as Fukui orbitals, chemical shifts, and Raman shifts can also be performed. This provides a computational means to probe electronic properties not readily accessible at scale using experimental techniques.

bioinformatics↗

bioRxiv2024

Gene PointNet for Tumor Classification

The rising incidence of cancer underscores the imperative for innovative diagnostic and prognostic methodologies. This study delves into the potential of RNA-Seq gene expression data to enhance cancer classification accuracy. Introducing a pioneering approach, we model gene expression data as point clouds, capitalizing on the datas intrinsic properties to bolster classification performance. Utilizing PointNet, a typical technique for processing point cloud data, as our frameworks cornerstone, we incorporate inductive biases pertinent to gene expression and pathways. This integration markedly elevates model efficacy, culminating in developing an end-to-end deep learning classifier with an accuracy rate surpassing 99%. Our findings not only illuminate the capabilities of AI-driven models in the realm of oncology but also highlight the criticality of acknowledging biological dataset nuances in model design. This research provides insights into application of deep learning in medical science, setting the stage for further innovation in cancer classification through sophisticated biological data analysis. The source code for our study is accessible at: https://github.com/cialab/GPNet.

bioinformatics↗

bioRxiv2023

Systems genetics uncover new loci containing functional gene candidates in Mycobacterium tuberculosis-infected Diversity Outbred mice.

Mycobacterium tuberculosis, the bacillus that causes tuberculosis (TB), infects 2 billion people across the globe, and results in 8-9 million new TB cases and 1-1.5 million deaths each year. Most patients have no known genetic basis that predisposes them to disease. We investigated the complex genetic basis of pulmonary TB by modelling human genetic diversity with the Diversity Outbred mouse population. When infected with M. tuberculosis, one-third develop early onset, rapidly progressive, necrotizing granulomas and succumb within 60 days. The remaining develop non-necrotizing granulomas and survive longer than 60 days. Genetic mapping using clinical indicators of disease, granuloma histopathological features, and immune response traits identified five new loci on mouse chromosomes 1, 2, 4, 16 and three previously identified loci on chromosomes 3 and 17. Quantitative trait loci (QTLs) on chromosomes 1, 16, and 17, associated with multiple correlated traits and had similar patterns of allele effects, suggesting these QTLs contain important genetic regulators of responses to M. tuberculosis. To narrow the list of candidate genes in QTLs, we used a machine learning strategy that integrated gene expression signatures from lungs of M. tuberculosis-infected Diversity Outbred mice with gene interaction networks, generating functional scores. The scores were then used to rank candidates for each mapped trait in each locus, resulting in 11 candidates: Ncf2, Fam20b, S100a8, S100a9, Itgb5, Fstl1, Zbtb20, Ddr1, Ier3, Vegfa, and Zfp318. Importantly, all 11 candidates have roles in infection, inflammation, cell migration, extracellular matrix remodeling, or intracellular signaling. Further, all candidates contain single nucleotide polymorphisms (SNPs), and some but not all SNPs were predicted to have deleterious consequences on protein functions. Multiple methods were used for validation including (i) a statistical method that showed Diversity Outbred mice carrying PWH/PhJ alleles on chromosome 17 QTL have shorter survival; (ii) quantification of S100A8 protein levels, confirming predicted allele effects; and (iii) infection of C57BL/6 mice deficient for the S100a8 gene. Overall, this work demonstrates that systems genetics using Diversity Outbred mice can identify new (and known) QTLs and new functionally relevant gene candidates that may be major regulators of granuloma necrosis and acute inflammation in pulmonary TB.

genetics↗

bioRxiv2023

Perivascular and peribronchiolar granuloma-associated lymphoid tissue and B-cell gene expression pathways identify asymptomatic Mycobacterium tuberculosis lung infection in Diversity Outbred mice{-}{-}{-}

Humans are highly genetically diverse, and most are resistant to Mycobacterium tuberculosis. However, lung tissue from genetically resistant humans is not readily available to identify potential mechanisms of resistance. To address this, we model M. tuberculosis infection in Diversity Outbred mice. Like humans, Diversity Outbred mice also exhibit genetically determined susceptibility to M. tuberculosis infection: Progressors who succumb within 60 days of a low dose aerosol infection due to acute necrotizing granulomas, and Controllers who maintain asymptomatic infection for at least 60 days, and then develop chronic pulmonary TB with occasional necrosis and cavitation, over months to greater than 1 year. Here, we identified specific regions of granuloma-associated lymphoid tissue (GrALT) and B-cell gene expression pathways as key features of asymptomatic lung infection using cytokine, antibody, granuloma image, and gene expression datasets. Cytokines and anti-M. tuberculosis cell wall antibodies discriminated acute vs chronic pulmonary TB but not asymptomatic lung infection. To find unique features of asymptomatic lung infection, we trained a weakly supervised, deep-learning neural network on lung histology images. The neural network accurately produced an interpretable imaging biomarker: perivascular and bronchiolar lymphocytic cuffs, a type of GrALT. We expected CD4 T cell genes would be highly expressed in asymptomatic lung infection. However, the significantly different, highly expressed genes in lungs of asymptomatically infected Diversity Outbred mice corresponded to B-cell activation, proliferation, and antigen-receptor signaling, including Fcrl1, Cd79, Pax5, Cr2, and Ms4a1. Overall, our results suggest that genetically controlled B-cell responses are important for establishing asymptomatic M. tuberculosis lung infection.

pathology↗