Aedin C. Culhane
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Aedin C. Culhane has a total of 822 co-authors. Here are the top co-authors by number of shared publications:
- John Quackenbush
Department of Biostatistics, T.H. Chan School of Public Health, Harvard University , Boston, MA 02115
19 shared publications - Benjamin Haibe-Kains
Princess Margaret Cancer Center, University Health Network
12 shared publications Eugen Dhimolea
Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, USA
7 shared publications- Levi Waldron
Institute for Implementation Science in Population Health, City University of New York School of Public Health, New York, NY
7 shared publications - William M. Gallagher
UCD School of Biomolecular and Biomedical Science, UCD Conway Institute, University College Dublin, D04 V1W8 Dublin, Ireland
7 shared publications - Chen Meng
Bavarian Center for Biomolecular Mass Spectrometry (BayBioMS), Technical University of Munich, Freising, Germany
6 shared publications
Latest Publications
- 1 September 2026
- Published by Oxford University Press (OUP) in JNCI: Journal of the National Cancer Institute
- Vol. 118 (10) , 1799-1801
Rising colorectal cancer mortality among young adults in many countries
- 1 September 2026
- Published by Oxford University Press (OUP) in JNCI: Journal of the National Cancer Institute
- Vol. 118 (10) , 1799-1801
Rising colorectal cancer mortality among young adults in many countries
- 1 September 2026
- Published by Springer Nature in Nature Biotechnology
- Vol. 44 (9) , 1433-1436
Ensuring multiomics data reproducibility for artificial intelligence with reference materials as a common calibrator
- 1 September 2026
- Published by Springer Nature in Nature Biotechnology
- Vol. 44 (9) , 1433-1436
Ensuring multiomics data reproducibility for artificial intelligence with reference materials as a common calibrator
- Open Access
- 1 August 2026
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (Supplement)
Semi-supervised learning for automated perineural invasion detection in multi-organ H&E whole slide images
Motivation: Perineural invasion (PNI) is an important pathological phenotype associated with poor prognosis in multiple malignancies. The primary detection method is visual inspection of whole slide images (WSIs), which is labor-intensive, time-consuming, subjective, and prone to high inter-observer variability. Developing reliable, accurate deep learning models for PNI detection is constrained by the lack of pixel-level annotated WSIs. Results: We evaluated four backbone architectures and two different approaches to improve PNI detection in a multi-organ dataset of colon, prostate, and pancreatic adenocarcinomas. We report three key findings. First, two pathology-pretrained foundation models, Virchow-2 and UNI, substantially outperformed ImageNet-pretrained CNNs (EfficientNet-B3, ConvNeXt-2), with distinct baseline error profiles reflecting differences in pretraining-data composition. Second, a data curation strategy driven by confidence-based pseudo-labelling (threshold P > .9) with human-in-the-loop review expanded the dataset from 262 to 352 WSIs, yielding a 12.4% relative F1 improvement (0.740 to 0.832) and a 55.5% reduction in false positives per slide; an ablation attributed 70% of the F1 gain to data volume and 41% of the FP reduction to benign-class curation. Third, per-organ analysis revealed that the primary driver of this improvement was not data volume alone but the targeted annotation enrichment of underrepresented morphologies in adjacent-normal and benign tissue, including desmoplastic stroma, crypts, and small blood vessels, that had been a systematic source of false positive predictions across tissue types. Availability: Our implementation is available at https://github.com/AhmadAlkhan/PNI_SSL.
- 1 August 2026
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (Supplement)
Open AccessSemi-supervised learning for automated perineural invasion detection in multi-organ H&E whole slide images
Motivation: Perineural invasion (PNI) is an important pathological phenotype associated with poor prognosis in multiple malignancies. The primary detection method is visual inspection of whole slide images (WSIs), which is labor-intensive, time-consuming, subjective, and prone to high inter-observer variability. Developing reliable, accurate deep learning models for PNI detection is constrained by the lack of pixel-level annotated WSIs. Results: We evaluated four backbone architectures and two different approaches to improve PNI detection in a multi-organ dataset of colon, prostate, and pancreatic adenocarcinomas. We report three key findings. First, two pathology-pretrained foundation models, Virchow-2 and UNI, substantially outperformed ImageNet-pretrained CNNs (EfficientNet-B3, ConvNeXt-2), with distinct baseline error profiles reflecting differences in pretraining-data composition. Second, a data curation strategy driven by confidence-based pseudo-labelling (threshold P > .9) with human-in-the-loop review expanded the dataset from 262 to 352 WSIs, yielding a 12.4% relative F1 improvement (0.740 to 0.832) and a 55.5% reduction in false positives per slide; an ablation attributed 70% of the F1 gain to data volume and 41% of the FP reduction to benign-class curation. Third, per-organ analysis revealed that the primary driver of this improvement was not data volume alone but the targeted annotation enrichment of underrepresented morphologies in adjacent-normal and benign tissue, including desmoplastic stroma, crypts, and small blood vessels, that had been a systematic source of false positive predictions across tissue types. Availability: Our implementation is available at https://github.com/AhmadAlkhan/PNI_SSL.
- Top Cited Papers
- 2 January 2026
- Published by Springer Nature in Nature Reviews Cancer
- Vol. 26 (3) , 217-229
Convergence of machine learning and genomics for precision oncology
The number of data points per patient considered at the point-of-care in precision cancer medicine continues to increase, and it is accompanied by a growing challenge of translating these observations into clinical insights. This is a time-intensive and laborious process for oncology professionals and molecular tumour boards. As large clinicogenomic datasets and data-sharing protocols mature alongside machine learning methods, molecular diagnostic workflows have an opportunity to integrate these tools. This integration can help extract more information from next-generation sequencing data, enhance cancer variant interpretation, streamline case review and generate therapeutic hypotheses for biomarker-negative patients at the point-of-care. Although machine learning holds promise for precision oncology, responsible implementation and model evaluation remain essential for clinical adoption.
- 2 January 2026
- Published by Springer Nature in Nature Reviews Cancer
- Vol. 26 (3) , 217-229
Top Cited PapersConvergence of machine learning and genomics for precision oncology
The number of data points per patient considered at the point-of-care in precision cancer medicine continues to increase, and it is accompanied by a growing challenge of translating these observations into clinical insights. This is a time-intensive and laborious process for oncology professionals and molecular tumour boards. As large clinicogenomic datasets and data-sharing protocols mature alongside machine learning methods, molecular diagnostic workflows have an opportunity to integrate these tools. This integration can help extract more information from next-generation sequencing data, enhance cancer variant interpretation, streamline case review and generate therapeutic hypotheses for biomarker-negative patients at the point-of-care. Although machine learning holds promise for precision oncology, responsible implementation and model evaluation remain essential for clinical adoption.
- Open Access
Enabling reproducible and reusable genetic demultiplexing benchmarking with Nextflow and Apptainer
Reproducibility, reusability, and portability were identified as key requirements for analysis pipelines in genomics. A previous study identified that less than 4% of notebooks used for biomedical research were fully reproducible. Reusable and portable software are required to ensure compatibility with federated and cloud computing. The field requires computational workflows that meet high standards for reproducibility, reusability, and portability to allow benchmarking to be repeated and updated periodically. This is of particular importance in areas where new methods are frequently released, such as demultiplexing scRNA-seq, a critical early step in most scRNA-seq analysis pipelines. To address this, we developed demux_bench for benchmarking genetic demultiplexing methods in single-cell RNA sequencing, which meets the gold standard for reproducibility of computational workflows, and incorporates best practices for reusability and portability. We used workflow manager Nextflow to enable the simulation and testing of various benchmarking scenarios and methods in parallel from a single pipeline execution. Different experimental configurations can be simulated, including % doublets and class size imbalance. The pipeline includes genotype-free methods, Vireo and souporcell, and additional existing or novel methods can be added modularly. Demux_bench is configurable to reproduce specific analyses and generalisable to address new research questions. Software dependencies are handled through containers via Apptainer, allowing portability to different compute environments and avoiding the need for manual installation of software. Demux_bench is available on WorkflowHub ( https://workflowhub.eu/workflows/1769 ), and can also be run on Galaxy and other platforms through RO-Crates. Demux_bench facilitates gold standard benchmarking for genetic demultiplexing through reproducibility, reusability, scalability, and portability.
Open AccessEnabling reproducible and reusable genetic demultiplexing benchmarking with Nextflow and Apptainer
Reproducibility, reusability, and portability were identified as key requirements for analysis pipelines in genomics. A previous study identified that less than 4% of notebooks used for biomedical research were fully reproducible. Reusable and portable software are required to ensure compatibility with federated and cloud computing. The field requires computational workflows that meet high standards for reproducibility, reusability, and portability to allow benchmarking to be repeated and updated periodically. This is of particular importance in areas where new methods are frequently released, such as demultiplexing scRNA-seq, a critical early step in most scRNA-seq analysis pipelines. To address this, we developed demux_bench for benchmarking genetic demultiplexing methods in single-cell RNA sequencing, which meets the gold standard for reproducibility of computational workflows, and incorporates best practices for reusability and portability. We used workflow manager Nextflow to enable the simulation and testing of various benchmarking scenarios and methods in parallel from a single pipeline execution. Different experimental configurations can be simulated, including % doublets and class size imbalance. The pipeline includes genotype-free methods, Vireo and souporcell, and additional existing or novel methods can be added modularly. Demux_bench is configurable to reproduce specific analyses and generalisable to address new research questions. Software dependencies are handled through containers via Apptainer, allowing portability to different compute environments and avoiding the need for manual installation of software. Demux_bench is available on WorkflowHub ( https://workflowhub.eu/workflows/1769 ), and can also be run on Galaxy and other platforms through RO-Crates. Demux_bench facilitates gold standard benchmarking for genetic demultiplexing through reproducibility, reusability, scalability, and portability.
- 3 May 2024
- Published by Springer Nature in Nature Genetics
- Vol. 56 (5) , 733-736
The 1+Million Genomes Minimal Dataset for Cancer
- 3 May 2024
- Published by Springer Nature in Nature Genetics
- Vol. 56 (5) , 733-736
The 1+Million Genomes Minimal Dataset for Cancer
- Open Access
demuxSNP: supervised demultiplexing scRNAseq using cell hashing and SNPs
Background Multiplexing single-cell RNA sequencing experiments reduces sequencing cost and facilitates larger scale studies. However, factors such as cell hashing quality and class size imbalance impact demultiplexing algorithm performance, reducing cost effectiveness Findings We propose a supervised algorithm, demuxSNP, leveraging both cell hashing and genetic variation between individuals (SNPs). The supervised algorithm addresses fundamental limitations in demultiplexing with only one data modality. The genetic variants (SNPs) of the subset of cells assigned with high confidence using a probabilistic hashing algorithm are used to train a KNN classifier that predicts the demultiplexing classes of unassigned or uncertain cells. We benchmark demuxSNP against hashing (HTODemux, cellhashR, GMM-demux, demuxmix) and genotype-free SNP (souporcell) methods on simulated and real data from renal cell cancer. Our results demonstrate that demuxSNP outperformed standalone hashing methods on low quality hashing data, improving overall classification accuracy and allowing more high RNA quality cells to be recovered. Through varying simulated doublet rates, we show genotype-free SNP methods are unable to identify biological samples with low cell counts at high doublet rates. When compared to unsupervised SNP demultiplexing methods, demuxSNP’s supervised approach was more robust to doublet rate in experiments with class size imbalance. Conclusions demuxSNP is a performant demultiplexing approach that uses hashing and SNP data to demultiplex datasets with low hashing quality where biological samples are genetically distinct. Unassigned cells (negatives) with high RNA quality can be recovered, making more cells available for analysis, especially when applied to data with low hashing quality or suspected misassigned cells. Pipelines for simulated data and processed benchmarking data for 5-50% doublets are publicly available. demuxSNP is available as an R/Bioconductor package (https://doi.org/doi:10.18129/B9.bioc.demuxSNP).
Open AccessdemuxSNP: supervised demultiplexing scRNAseq using cell hashing and SNPs
Background Multiplexing single-cell RNA sequencing experiments reduces sequencing cost and facilitates larger scale studies. However, factors such as cell hashing quality and class size imbalance impact demultiplexing algorithm performance, reducing cost effectiveness Findings We propose a supervised algorithm, demuxSNP, leveraging both cell hashing and genetic variation between individuals (SNPs). The supervised algorithm addresses fundamental limitations in demultiplexing with only one data modality. The genetic variants (SNPs) of the subset of cells assigned with high confidence using a probabilistic hashing algorithm are used to train a KNN classifier that predicts the demultiplexing classes of unassigned or uncertain cells. We benchmark demuxSNP against hashing (HTODemux, cellhashR, GMM-demux, demuxmix) and genotype-free SNP (souporcell) methods on simulated and real data from renal cell cancer. Our results demonstrate that demuxSNP outperformed standalone hashing methods on low quality hashing data, improving overall classification accuracy and allowing more high RNA quality cells to be recovered. Through varying simulated doublet rates, we show genotype-free SNP methods are unable to identify biological samples with low cell counts at high doublet rates. When compared to unsupervised SNP demultiplexing methods, demuxSNP’s supervised approach was more robust to doublet rate in experiments with class size imbalance. Conclusions demuxSNP is a performant demultiplexing approach that uses hashing and SNP data to demultiplex datasets with low hashing quality where biological samples are genetically distinct. Unassigned cells (negatives) with high RNA quality can be recovered, making more cells available for analysis, especially when applied to data with low hashing quality or suspected misassigned cells. Pipelines for simulated data and processed benchmarking data for 5-50% doublets are publicly available. demuxSNP is available as an R/Bioconductor package (https://doi.org/doi:10.18129/B9.bioc.demuxSNP).
- Open Access
- 7 January 2024
- Published by Adjacent Digital Politics Ltd in Open Access Government
- Vol. 43 (1) , 94-95
From selfish silo to collaborative culture – embracing data-enabled cancer research
From selfish silo to collaborative culture – embracing data-enabled cancer research: Aedin Culhane and Mark Lawler, Co-Leads of the eHealth Hub for Cancer, reflect on their data-enabled cancer research journeys, how their collaborative team science approach has reaped significant dividends in cancer research and policy and how the hub is inducing a paradigm shift in how health data are deployed on the island of Ireland. An article in The Wall Street Journal (not the normal reading material for scientists) in 2011 highlighted a new approach to performing scientific research that was gaining significant credence at the time. Entitled ‘The New Einsteins Will Be Scientists Who Share’ (with the strapline ‘From cancer to cosmology, researchers could race ahead by working together – online and in the open’), the article presaged an unprecedented change in how scientists interact with each other, ushering in a culture of collaboration and cross-disciplinary research. Nowhere was this change more obvious than in the genomics and data science community, where a bottom-up movement led to the creation of the Global Alliance for Genomics and Health (GA4GH), bringing together researchers from around the world to work together to address some of human health’s greatest challenges through the deployment of data and data tools.
- 7 January 2024
- Published by Adjacent Digital Politics Ltd in Open Access Government
- Vol. 43 (1) , 94-95
Open AccessFrom selfish silo to collaborative culture – embracing data-enabled cancer research
From selfish silo to collaborative culture – embracing data-enabled cancer research: Aedin Culhane and Mark Lawler, Co-Leads of the eHealth Hub for Cancer, reflect on their data-enabled cancer research journeys, how their collaborative team science approach has reaped significant dividends in cancer research and policy and how the hub is inducing a paradigm shift in how health data are deployed on the island of Ireland. An article in The Wall Street Journal (not the normal reading material for scientists) in 2011 highlighted a new approach to performing scientific research that was gaining significant credence at the time. Entitled ‘The New Einsteins Will Be Scientists Who Share’ (with the strapline ‘From cancer to cosmology, researchers could race ahead by working together – online and in the open’), the article presaged an unprecedented change in how scientists interact with each other, ushering in a culture of collaboration and cross-disciplinary research. Nowhere was this change more obvious than in the genomics and data science community, where a bottom-up movement led to the creation of the Global Alliance for Genomics and Health (GA4GH), bringing together researchers from around the world to work together to address some of human health’s greatest challenges through the deployment of data and data tools.
- Open Access
- 2 January 2024
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 13
demuxSNP: supervised demultiplexing single-cell RNA sequencing using cell hashing and SNPs
Background: Multiplexing single-cell RNA sequencing experiments reduces sequencing cost and facilitates larger-scale studies. However, factors such as cell hashing quality and class size imbalance impact demultiplexing algorithm performance, reducing cost-effectiveness. Findings: We propose a supervised algorithm, demuxSNP, which leverages both cell hashing and genetic variation between individuals (single-nucletotide polymorphisms [SNPs]). demuxSNP addresses fundamental limitations in demultiplexing methods that use only one data modality. Some cells may be confidently demultiplexed using probabilistic hashing methods. demuxSNP uses these data to infer the genotype of singlet and doublet clusters and predict on cells assigned as negative, uncertain, or doublet using a nearest-neighbor approach adapted for missing data. We benchmarked demuxSNP against hashing, genotype-free SNP and hybrid methods on simulated and real data from renal cell cancer. demuxSNP outperformed standalone hashing methods on low-quality hashing data benchmark, improved overall classification accuracy, and allowed more high RNA quality cells to be recovered. Through varying simulated doublet rates, we showed that genotype-free SNP and hybrid methods that leverage them were impacted by class size imbalance and doublet rate. demuxSNP’s supervised approach was more robust to doublet rate in experiments with class size imbalance. Conclusions: demuxSNP uses hashing and SNP data to demultiplex datasets with low hashing quality where biological samples are genetically distinct. Unassigned or negative cells with high RNA quality are recovered, making more cells available for analysis. Data simulation and benchmarking pipelines as well as processed benchmarking data for 5–50% doublets are publicly available. demuxSNP is available as an R/Bioconductor package (https://doi.org/doi:10.18129/B9.bioc.demuxSNP).
- 2 January 2024
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 13
Open AccessdemuxSNP: supervised demultiplexing single-cell RNA sequencing using cell hashing and SNPs
Background: Multiplexing single-cell RNA sequencing experiments reduces sequencing cost and facilitates larger-scale studies. However, factors such as cell hashing quality and class size imbalance impact demultiplexing algorithm performance, reducing cost-effectiveness. Findings: We propose a supervised algorithm, demuxSNP, which leverages both cell hashing and genetic variation between individuals (single-nucletotide polymorphisms [SNPs]). demuxSNP addresses fundamental limitations in demultiplexing methods that use only one data modality. Some cells may be confidently demultiplexed using probabilistic hashing methods. demuxSNP uses these data to infer the genotype of singlet and doublet clusters and predict on cells assigned as negative, uncertain, or doublet using a nearest-neighbor approach adapted for missing data. We benchmarked demuxSNP against hashing, genotype-free SNP and hybrid methods on simulated and real data from renal cell cancer. demuxSNP outperformed standalone hashing methods on low-quality hashing data benchmark, improved overall classification accuracy, and allowed more high RNA quality cells to be recovered. Through varying simulated doublet rates, we showed that genotype-free SNP and hybrid methods that leverage them were impacted by class size imbalance and doublet rate. demuxSNP’s supervised approach was more robust to doublet rate in experiments with class size imbalance. Conclusions: demuxSNP uses hashing and SNP data to demultiplex datasets with low hashing quality where biological samples are genetically distinct. Unassigned or negative cells with high RNA quality are recovered, making more cells available for analysis. Data simulation and benchmarking pipelines as well as processed benchmarking data for 5–50% doublets are publicly available. demuxSNP is available as an R/Bioconductor package (https://doi.org/doi:10.18129/B9.bioc.demuxSNP).
- Open Access
- 14 October 2023
- Published by Elsevier in Journal of Cancer Policy
- Vol. 38, 100448
Health diplomacy in action: The cancer legacy of the Good Friday Agreement
- 14 October 2023
- Published by Elsevier in Journal of Cancer Policy
- Vol. 38, 100448
Open AccessHealth diplomacy in action: The cancer legacy of the Good Friday Agreement