Biology subjects

Hernando, B.

Publications and source records attributed to Hernando, B..

6 recordsLinked to original sources

bioRxiv2026

Explainable machine learning relates histological to genomic pathology

Background & AimsHaematoxylin and eosin (H&E) staining remains the diagnostic gold standard for solid cancers, including hepatocellular carcinoma, and is increasingly complemented by genomic profiling for precision medicine. Inferring genomic alterations directly from H&E images could streamline testing, but heterogeneity and biases in human training data limit interpretation of genotype-phenotype associations. Here, we aimed to relate histologic to genomic pathology to provide biological explainability for mutation prediction models and assess the impact of germline variation on model performance. MethodsWe analysed 597 murine liver tumours with matched whole-genome sequencing and histopathology (163,835 image tiles; 22.9 million nuclei). Our controlled in vivo design accounted for germline variation, biological sex, and causal mutagen (N-diethylnitrosamine), removing confounding factors present in human cohorts. We trained and evaluated deep learning and supervised machine learning models to predict germline variation and cancer driver alterations from H&E. ResultsModelling accurately predicted germline and somatic alterations from histology, at both locus-specific and genome-wide scales. Quantitative image analysis revealed an unexpected association between Egfr driver mutations and hepatic steatosis, linking genotype to an interpretable morphological phenotype. While model performance declined when applied to tumours from unrepresented genetic backgrounds, this limitation was biologically informative, revealing strain-dependent differences in tumour evolution, notably the prevalence of whole-genome duplication. ConclusionsMachine learning integration of histological and genomic pathology enables accurate, interpretable inference of genetic alterations from H&E, potentially reducing reliance on costly ancillary molecular assays. Our predictions are supported by human-interpretable biological features, addressing concerns around "black-box" technologies. However, caution is required when applying such methods to samples with a genetic background that, even if closely related, is beyond the genetic horizon of training data.

pathology↗

bioRxiv2026

WNT-driven chromosomal instability as a biomarker for PORCN inhibition

Targeting Porcupine (PORCN), a key regulator of the WNT-signalling pathway, has shown therapeutic potential in multiple cancers. Despite strong target engagement and acceptable safety profiles through human phase I clinical trials, low phase II efficacy has stalled further clinical development. Given that aberrant WNT signalling can drive tumorigenesis by inducing chromosomal instability (CIN), we hypothesised that genomic CIN signatures might serve as a predictive biomarker to help improve response rates. Using a controlled in vitro model and single-cell whole-genome sequencing, we demonstrate that acute WNT-activation directly induces three distinct types of CIN: whole genome duplication, replication stress, and impaired homologous recombination. We translated these observations into a composite CIN signature biomarker that significantly correlated with both genetic dependency and pharmacological inhibition of PORCN across 195 and 24 cell lines, respectively. Through a large-scale meta-analysis of patient-derived and cell line xenografts, we established that this composite CIN signature biomarker quantitatively predicts in vivo PORCN inhibitor sensitivity (R=-0.71, p<0.002). By applying an optimised biomarker threshold, refined through modelling of human patient data, to the The Cancer Genome Atlas dataset, we successfully retrospectively modelled previous trial results and identified gastroesophageal cancers as a high-prevalence (36.6%) indication for future development. We validated this strategy in a mouse clinical trial of gastric and esophageal xenografts, where biomarker-guided stratification achieved an objective response rate of 60% and significantly decreased risk of progression (HR=0.21, p=0.0345). These data establish an actionable, trail-ready framework for further PORCN inhibitor clinical development.

cancer biology↗

bioRxiv2025

Tracking ongoing chromosomal instability using single-cell whole-genome sequencing

Chromosomal instability (CIN) generates aneuploid genomes that are characteristic of most cancers. While bulk genome sequencing reveals historical CIN, it lacks the resolution to identify ongoing CIN that actively shapes genome evolution. Here, we present a computational framework that leverages single-cell whole-genome sequencing (scWGS) to identify and quantify ongoing CIN by detecting cell-unique copy number alterations and probabilistically mapping them to known CIN signatures. We validated this framework generating in vitro models with four types of induced CIN, correctly identifying the induced-CIN type in each case. When applied to cell lines and organoids with ongoing homologous recombination deficiency, our method showed improved identification of sensitivity to PARP inhibition and platinum-based chemotherapy. Analysing scWGS data from 8 triple-negative breast cancers, we linked ongoing impaired non-homologous end joining to subclonal diversification, a finding further supported in cohorts of 179 unmatched primary and metastatic TNBCs and 39 matched cases. Collectively, our results demonstrate that distinguishing ongoing from historical chromosomal instability uncovers a distinct dimension of tumour evolution, suggesting that effective precision oncology will require integrating measurements of both past genomic scars and active mutational processes.

genomics↗

bioRxiv2025

Forecasting oncogene amplification and tumour suppressor deletion

Oncogene amplification and tumour suppressor deletion can drive tumour initiation, progression and treatment resistance. Detection at diagnosis often signals poor prognosis, but it can also enable opportunities for treatment with highly effective targeted therapies. Predicting the likelihood that a patient will acquire these driver alterations in the future using a genomic test represents an opportunity to realise the benefits of interventions earlier, potentially with preventative intent. Here, we present a forecasting framework that takes as input a DNA copy number profile and predicts whether the tumour will acquire an oncogene amplification or tumour suppressor deletion in the future. This framework leverages mutation rate estimates from the input tumour, alongside gene-specific selection coefficients derived from a large cohort of 7,880 tumours. We demonstrate feasibility using 7,042 single-time-point samples and longitudinally collected tumour pairs from 44 prostate and 100 lung cancers, identifying tumours that went on to acquire amplifications at a later time point with an average AUC of 0.87. We show potential clinical utility by forecasting poor prognosis in low-grade gliomas via CDK4/PDGFRA amplification or CDKN2A deletion, and osimertinib resistance in lung cancers via MET amplification. This study serves as a proof-of-concept for a new class of biomarker, wherein selective pressures and mutation-generating processes can be harnessed to anticipate future genomic alterations.

genomics↗

bioRxiv2022

High-grade serous ovarian carcinoma organoids as models of chromosomal instability

High-grade serous ovarian carcinoma (HGSOC) is the most genomically complex cancer, characterised by ubiquitous TP53 mutation, profound chromosomal instability and heterogeneity. The mutational processes driving chromosomal instability in HGSOC can be distinguished by specific copy number signatures. To develop clinically relevant models of these mutational processes we derived 15 continuous HGSOC patient-derived organoids (PDOs). We carried out detailed bulk transcriptomic, bulk genomic, single cell genomic, and drug sensitivity characterisation of the organoids. We show that PDOs comprise communities of different clonal populations and represent models of different causes of chromosomal instability including homologous recombination deficiency, chromothripsis, tandem-duplicator phenotype and whole genome duplication. We also show that these PDOs can be used as exploratory tools to study transcriptional effects of copy number alterations as well as compound-sensitivity tests. In summary, HGSOC PDO cultures provide a genomic tool for studies of specific mutational processes and precision therapeutics.

cancer biology↗

bioRxiv2021

CNpare: matching DNA copy number profiles

Selecting the optimal cancer cell line for an experiment can be challenging given the diversity of lines available. Cell lines are often chosen based on their tissue of origin, however, the results of large-scale pan-cancer studies suggest that matching lines based on molecular features may be more appropriate. Existing approaches are available for matching lines based on gene expression, DNA methylation or low resolution DNA copy number features. However, a specific tool for computing similarity based on high resolution genome-wide copy number profiles is lacking. Here, we present CNpare, which identifies similar cell line models based on genome-wide DNA copy number. CNpare compares copy number profiles using four different similarity metrics, quantifies the extent of genome differences between pairs, and facilitates comparison based on copy number signatures. CNpare incorporates a precomputed database of 1,170 human cancer cell line profiles for comparison. In an analysis of separate cultures of 304 cell line pairs, CNpare identified the matched lines in all cases. CNpare provides a powerful solution to the problem of selecting the best cell line models for cancer research, especially in the context of studying chromosomal instability.

bioinformatics↗