Biology subjects

Minhas, F.

Publications and source records attributed to Minhas, F..

10 recordsLinked to original sources

bioRxiv2026

Discovering Prognostic Risk Strata and Heterogeneous Progression Routes from Cross-Sectional Survival Data

Inferring how individual patients progress towards higher disease risk typically requires longitudinal observations, which are often unavailable. Population-level analyses of cross-sectional data can identify differences between risk groups but may obscure heterogeneous progression patterns across individual patients. In this work, we introduce GOTSurv, a framework for patient-specific progression modelling. Using cross-sectional survival data, it first learns to rank patients by survival risk and automatically stratify them into prognostic risk groups without requiring predefined risk cut-offs. It then uses unbalanced optimal transport to infer patient-specific changes in the magnitude and direction of features associated with transitions towards higher-risk strata. We evaluated GOTSurv across ten TCGA cancer cohorts using two modalities: gene expression (n=4,767) and whole-slide image representations (n=4,655), with external validation in two independent cohorts. Compared with established survival models, GOTSurv generally achieved stronger prognostic stratification without requiring predefined risk cut-offs, while retaining competitive survival-ranking accuracy. More importantly, GOTSurv enabled the discovery of individualised progression trajectories associated with increasing risk, revealing distinct progression patterns even among patients within the same risk stratum. Overall, GOTSurv provides a novel framework for inferring heterogeneous, patient-specific progression patterns from cross-sectional survival data, enabling the identification of distinct pathways associated with increasing disease risk and informing the development of personalised therapeutic strategies. The code for GOTSurv is available at https://github.com/EtharZaid/GOTSurv/.

bioinformatics↗

bioRxiv2026

Whole slide image analysis of the endometrial decidual reaction reveals multiscale perturbations associated with miscarriage

The inflammatory decidual reaction renders the cycling endometrium transiently permissive for embryo implantation before transforming it into the decidua, the maternal bed accommodating the fetal placenta during pregnancy. Disruptions in decidual tissue remodeling are linked to miscarriage and other pregnancy disorders. However, endometrial assessment is hampered by a lack of affordable technologies capable of mapping the spatiotemporal dysregulation of this dynamic and complex tissue. Employing a graph neural network on whole slide images of 493 CD56-immunostained endometrial samples, Endometronome was developed as a deep learning tool to spatially track the decidual reaction and provide accurate estimates of marker gene expression. When applied to 2,690 additional biopsies, this model consistently identified morphological correlates of prior miscarriage burden, a proxy for future risk. Further, a morphological signature indicative of metabolic glandular impairment discriminated between clinical miscarriage presentations. These findings illustrate how advanced imaging analysis of routine histology can transform miscarriage prevention strategies.

pathology↗

bioRxiv2026

DCAFA: Differential Community Abundance and Feature Analysis for Histological Images

Histological images are composed of diverse local structures such as cells, glands, or tissue patches, whose organisation and relative abundance reflect underlying biological processes. While most computational approaches focus on analysing individual features, many clinically relevant signals arise from changes in the composition of these structures rather than isolated measurements. This motivates the need for explicitly modelling how groups of similar instances vary across samples and relate to outcomes. We present DCAFA (Differential Community Abundance and Feature Attribution Analysis), a regression-based framework for analysing hierarchical biomedical data through both compositional and feature-level perspectives. DCAFA groups instances into latent communities representing recurring morphological or phenotypic patterns, and then performs two complementary analyses: (i) community composition analysis, which identifies groups that are enriched or depleted across outcomes, and (ii) feature attribution analysis, which quantifies how instance-level features relate to outcomes directly or within specific communities. Both use generalised linear and mixed-effects models, enabling covariate adjustment and inference through effect sizes, confidence intervals, and false discovery rate control. We demonstrate the utility of DCAFA across multiple biomedical settings, including endometrial histopathology, spatial transcriptomics, multiplex immunofluorescence imaging, and predefined cell-type analyses in colorectal cancer. These examples identify interpretable compositional shifts and context-specific feature associations that are not captured by conventional feature-based approaches. By unifying differential abundance testing and feature attribution within a single statistical framework, DCAFA serves as an openly available toolbox that provides a practical and interpretable means of linking tissue composition with clinical or molecular outcomes in biomedical imaging data. Code available at: https://github.com/wgrgwrght/DCAFA

bioinformatics↗

bioRxiv2026

GOTFlow: Learning Directed Population Transitions from Cross-Sectional Biomedical Data with Optimal Transport

MotivationMany biological and clinical processes are dynamic, yet most datasets are cross-sectional, capturing populations at discrete states rather than tracking individuals over time. This makes it difficult to quantify how populations change across developmental, physiological, or disease-associated conditions. Existing trajectory and transport-based methods often rely on fixed feature spaces, assumptions tailored to transcriptomic time-course data, or approximately linear progression, limiting their ability to model heterogeneous and unbalanced transitions across diverse biomedical modalities. Flexible methods are needed that can infer directed population-level change from cross-sectional data while retaining biological interpretability. ResultsWe present GOTFlow, a framework for learning directed population transitions from cross-sectional biomedical data using graph-constrained optimal transport in a learned latent space. GOTFlow integrates representation learning with unbalanced optimal transport to jointly estimate embeddings and transport couplings between biological states. This enables hypothesis-driven modelling of progression structures while accommodating non-linear geometry, branching relationships, and changes in population mass. From the inferred transport plans, GOTFlow derives interpretable summaries of dynamics, including drift vectors quantifying transitions, and feature-level transported changes that highlight molecular drivers of progression. In synthetic data, GOTFlow recovered known transitions with strong agreement between inferred and ground-truth drifts. Across three biological applications, endometrial remodelling, breast cancer risk progression, and prion disease, GOTFlow identified state-to-state transitions and biologically meaningful feature shifts reflecting impaired decidualisation, increasing cancer risk, and neurodegenerative progression. These results establish GOTFlow as a general and interpretable framework for analysing directed population dynamics from cross-sectional data. AvailabilityCode available at: https://github.com/wgrgwrght/GOTFlow

bioinformatics↗

bioRxiv2025

Star-Motifs: Revealing Single-Cell Spatiotypes from Routine Histology Using Star-Convex Neighborhoods

The tumor microenvironment (TME) is a dynamic interplay among cancer, immune, and stromal cells that profoundly influences tumor growth, progression, and treatment response. Although recent spatial transcriptomics and multiplex imaging technologies offer fine-grained insights into the TME, their high cost and lengthy turnaround times limit routine clinical use. In contrast, whole slide images (WSIs) of Hematoxylin and eosin (H&E)-stained are readily available for most patients and provide valuable information about tumor biology. Here, we present a method that explicitly models each cells local neighborhood to identify recurrent spatial patterns, which we call "Star-Motifs." We first encode the local microenvironment of each cell using a "Star-Environment" representation that captures distances and angles to multiple surrounding cell types. We then perform unsupervised clustering of these descriptors to discover distinct Star-Motifs. We validated our method on 1.8 billion cells across 3,105 diagnostic slides from The Cancer Genome Atlas, spanning five distinct cell types: neoplastic, inflammatory, stromal, necrotic, and non-neoplastic. Our method uncovered a diverse range of Star-Motifs, from densely packed tumor clusters to tumor cells surrounded by immune cells (tumor-infiltrating lymphocytes). At the patient level, the relative abundances of these Star-Motifs correlate with patient survival outcomes, immune subtypes, and molecular alterations, offering interpretable, clinically relevant insights. Moreover, classical machine learning models trained on these abundances match or surpass deep learning approaches in predicting overall survival and immune-related biomarkers while remaining interpretable. By capturing higher-order spatial arrangements from routinely available histology slides, Star-Motifs provides a scalable, transparent platform for TME profiling, biomarker discovery, and personalized risk assessment in oncology.

cancer biology↗

bioRxiv2024

Predicting Degradation Potential of Protein Targeting Chimeras

PRoteolysis TArgeting Chimeras (PROTACs) can inhibit protein activity by utilizing natural proteasomal degradation pathways for the degradation of target proteins. Being able to determine the degradation potential of PROTACs is crucial in drug development as it can lead to time, labor and cost savings. In this paper, we present a novel machine-learning pipeline that utilizes common compound fingerprints and a pre-trained graph neural network for the prediction of half-maximal degradation concentration of PROTACs by benchmarking a variety of protein tertiary structures and chemical features. Based on critical analysis of our cross-validation and independent test results, we have highlighted several key challenges underlying this prediction problem that need to be addressed to improve the generalization of predictive models in this domain. Moreover, we demonstrate the effectiveness of our approach by testing it on two different datasets and show that it performs better than the current state of the art with an AUC-ROC of 0.85 and accuracy of 0.875 on the DeepPROTACs test dataset.

bioinformatics↗

bioRxiv2024

Predicting small-molecule inhibition of protein complexes

MotivationProtein-Protein Interactions (PPIs) are crucial in biological processes and disease mechanisms, underscoring the importance of discovering PPI inhibitors in drug development. Machine learning can expedite this discovery process. Although machine learning techniques for predicting general compound inhibition are available, we are not aware of any that accurately forecast the inhibitory effect of a compound on a specific protein complex, utilizing inputs from both the compound and the protein complex. MethodsWe present the first targeted machine learning based predictor of small molecule based inhibition of protein complexes. Our proposed graph neural network integrates the structure of a protein complex, its protein-protein binding site or interface features and a compounds SMILES representation to predict the potential of the given compound to inhibit the interaction between proteins in the given complex in a targeted manner. ResultsValidated on the 2p2i-DB-v2 database, encompassing 714 inhibitors across 23 complexes with over 12,000 instances, our model achieves superior predictive accuracy (cross-validation AUC-ROC of 0.86), outperforming established kernel methods and pre-trained neural networks. We further tested the predictive performance of our model on two independent external datasets - one collected from recent publications and another consisting of putative inhibitors of the SARS-CoV-2-Spike and Human-ACE2 protein complex with AUC-ROCs of 0.82 and 0.78, respectively. Our targeted predictor introduces a novel approach for PPI inhibitor discovery, laying foundational work for future advancements in addressing this complex and previously unexplored prediction challenge. AvailabilityCode/supplementary material available: https://github.com/adibayaseen/PPI-Inhibitors

bioinformatics↗

bioRxiv2024

Buyer Beware: confounding factors and biases abound when predicting omics-based biomarkers from histological images

BackgroundRecent advancements in computational pathology have introduced deep learning methods to predict genomic, transcriptomic and molecular biomarkers from routine histology whole slide images (WSIs) for cancer diagnosis, prognosis, and treatment. However, existing methods often overlook the critical role of co-dependencies among biomarker statuses during training and inference. We hypothesize that this oversight results in models that predict the combined effect of multiple interdependent biomarkers rather than individual statuses independently, akin to attributing the quality of an orchestral symphony to a single instrument, highlighting limitations of current predictors. MethodsUsing large datasets (n = 8,221 patients), we conducted statistical co-dependence testing to demonstrate significant interdependencies among biomarker statuses in training datasets. Following standard protocols, we trained two machine learning models to predict biomarkers from WSIs achieving or matching state-of-the-art predictive performance. We then employed permutation testing and stratification analysis to evaluate their predictive quality based on the principle of conditional independence, i.e., if a model accurately captures the phenotypic influence of a specific biomarker independent of other biomarkers, its performance should remain consistent across subgroups of patients stratified by other biomarkers, aligning with its overall performance on the entire dataset. FindingsOur statistical analysis reveals significant interdependencies among biomarkers, reflecting expected co-occurrence and mutual exclusivity patterns influenced by pathological and biological processes that are consistent across datasets, as well as sampling artefacts that can be different across datasets. Our results indicate that the predictive quality of an image-based predictor for a biomarker is contingent on the status of other biomarkers, revealing that models capture aggregated influences rather than predicting individual statuses independently. For example, mutation predictions are confounded by the overall tumour mutation burden. We also show that, due to the presence of such correlations, deep learning models may not offer significant advantages in predicting certain biomarkers in comparison to simply using pathologist-assigned grades for their prediction. InterpretationWe show that current deep learning models in computational pathology fall short in isolating individual biomarker effects, leading to confounded and less precise predictions. Our findings suggest revisiting model training protocols to recognize and adjust for biomarker interdependencies at all development stages--from problem definition to usage guidelines. This involves selecting diverse datasets to reflect clinical heterogeneity, defining prediction variables or grouping patients based on co-dependencies, designing models to disentangle complex relationships, and stringent stratification testing. Clinically, failure to account for interdependencies may lead to suboptimal decisions, necessitating appropriate usage guidelines for predictive models.

cancer biology↗

bioRxiv2021

TIAToolbox: An End-to-End Toolbox for Advanced Tissue Image Analytics

Computational Pathology (CPath) has seen rapid growth in recent years, driven by advanced deep learning (DL) algorithms. These algorithms typically share the same sequence of steps. However, due to the sheer size and complexity of handling large multi-gigapixel whole-slide images, there is no open-source software library that provides a generic end-to-end API for pathology image analysis using best practices for CPath. Most researchers have designed custom pipelines from the bottom-up, restricting the development of advanced CPath algorithms to specialist users. To help overcome this bottleneck, we present TIAToolbox, a Python toolbox designed to make CPath more accessible to new and advanced CPath scientists and pathologists alike. We provide a usable and adaptable library with efficient, cutting-edge and unit-tested tools for data loading, pre-processing, model inference, post-processing and visualization. This enables all kinds of users to easily build upon recent DL developments in the CPath literature. TIAToolbox provides a user-friendly modular API to enable seamless integration of advanced DL algorithms. We show with the help of examples how state-of-the-art DL algorithms can be streamlined using TIAToolbox.

bioengineering↗

bioRxiv2020

Improving COVID-19 Testing Efficiency using Guided Agglomerative Sampling

One of the challenges in the current COVID-19 crisis is the time and cost of performing tests especially for large-scale population surveillance. Since, the probability of testing positive in large population studies is expected to be small (<15%), therefore, most of the test outcomes will be negative. Here, we propose the use of agglomerative sampling which can prune out multiple negative cases in a single test by intelligently combining samples from different individuals. The proposed scheme builds on the assumption that samples from the population may not be independent of each other. Our simulation results show that the proposed sampling strategy can significantly increase testing capacity under resource constraints: on average, a saving of ~40% tests can be expected assuming a positive test probability of 10% across the given samples. The proposed scheme can also be used in conjunction with heuristic or Machine Learning guided clustering for improving the efficiency of large-scale testing further. The code for generating the simulation results for this work is available here: https://github.com/foxtrotmike/AS.

bioinformatics↗