Tumour purity assessment with deep learning in colorectal cancer and impact on molecular analysis

Lydia A Schoenpflug; Aikaterini Chatzipli; Korsuk Sirinukunwattana; Susan Richman; Andrew Blake; James Robineau; Kirsten D Mertz; Clare Verrill; Simon J Leedham; Claire Hardy; Celina Whalley; Keara Redmond; Philip Dunne; Steven Walker; Andrew D Beggs; Ultan McDermott; Graeme I Murray; Leslie M Samuel; Matthew Seymour; Ian Tomlinson; Philip Quirke; S:CORT consortium; Jens Rittscher; Tim Maughan; Enric Domingo; Viktor H Koelzer

doi:10.1002/path.6376

Tumour purity assessment with deep learning in colorectal cancer and impact on molecular analysis

Lydia A Schoenpflug, Aikaterini Chatzipli, Korsuk Sirinukunwattana, Susan Richman, Andrew Blake, James Robineau, Kirsten D Mertz, Clare Verrill, Simon J Leedham, Claire Hardy, Celina Whalley, Keara Redmond, Philip Dunne, Steven Walker, Andrew D Beggs, Ultan McDermott, Graeme I Murray, Leslie M Samuel, Matthew Seymour, Ian TomlinsonPhilip Quirke, S:CORT consortium, Jens Rittscher, Tim Maughan^* (Corresponding Author), Enric Domingo^* (Corresponding Author), Viktor H Koelzer^* (Corresponding Author)

^*Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

1 Downloads (Pure)

Abstract

Abstract Tumour content plays a pivotal role in directing the bioinformatic analysis of molecular profiles such as copy number variation (CNV). In clinical application, tumour purity estimation (TPE) is achieved either through visual pathological review [conventional pathology (CP)] or the deconvolution of molecular data. While CP provides a direct measurement, it demonstrates modest reproducibility and lacks standardisation. Conversely, deconvolution methods offer an indirect assessment with uncertain accuracy, underscoring the necessity for innovative approaches. SoftCTM is an open-source, multiorgan deep-learning (DL) model for the detection of tumour and non-tumour cells in H&E-stained slides, developed within the Overlapped Cell on Tissue Dataset for Histopathology (OCELOT) Challenge 2023. Here, using three large multicentre colorectal cancer (CRC) cohorts (N?=?1,097 patients) with digital pathology and multi-omic data, we compare the utility and accuracy of TPE with SoftCTM versus CP and bioinformatic deconvolution methods (RNA expression, DNA methylation) for downstream molecular analysis, including CNV profiling. SoftCTM showed technical repeatability when applied twice on the same slide (r?=?1.0) and excellent correlations in paired H&E slides (r?>?0.9). TPEs profiled by SoftCTM correlated highly with RNA expression (r?=?0.59) and DNA methylation (r?=?0.40), while TPEs by CP showed a lower correlation with RNA expression (r?=?0.41) and DNA methylation (r?=?0.29). We show that CP and deconvolution methods respectively underestimate and overestimate tumour content compared to SoftCTM, resulting in 6?13% differing CNV calls. In summary, TPE with SoftCTM enables reproducibility, automation, and standardisation at single-cell resolution. SoftCTM estimates (M?=?58.9%, SD ±16.3%) reconcile the overestimation by molecular data extrapolation (RNA expression: M?=?79.2%, SD ±10.5, DNA methylation: M?=?62.7%, SD ±11.8%) and underestimation by CP (M?=?35.9%, SD ±13.1%), providing a more reliable middle ground. A fully integrated computational pathology solution could therefore be used to improve downstream molecular analyses for research and clinics. ? 2024 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Original language	English
Pages (from-to)	184-197
Number of pages	14
Journal	The Journal of pathology
Volume	265
Issue number	2
Early online date	22 Dec 2024
DOIs	https://doi.org/10.1002/path.6376
Publication status	E-pub ahead of print - 22 Dec 2024

Bibliographical note

The authors thank Claire Butler and Michael Youdell for excellent management in S:CORT and the MRC Clinical Trials Unit which provided the clinical data from the FOCUS trial with permission from the FOCUS trial steering group. We would further like to thank Indica Labs for providing the HALO™ software. The results published or shown here are based in part upon data generated by the TCGA Research Network established by the National Cancer Institute and National Human Genome Research Institute. Information about TCGA and the investigators and institutions who constitute the TCGA research network can be found at http://cancergenome.nih.gov. We would especially like to thank all patients who consented to take part in S:CORT and TCGA. The views expressed are those of the author(s) and not necessarily those of the National Health Service, the National Institute for Health and Care Research, or the Department of Health.

Data Availability Statement

FOCUS raw expression data and molecular metadata are publicly available at GEO under reference GSE156915. The transcriptomic data from GRAMPIAN are publicly available at the following link: https://www.scort.org/sites/default/files/exports/scort_ws3_grampian_export_84m9fndk/ws3_grampian_expression_raw.zip. Sequencing data from whole S:CORT are publicly available in EGA (EGAS00001001521). Additional S:CORT data are available to all academic researchers on submission of a data request to the data access committee. For commercial agencies, the data will be made available through Cancer Research Horizons acting on behalf of the funders and consortium members. The TCGA datasets and images analysed in this study are openly and publicly available at https://portal.gdc.cancer.gov/.

Keywords

pathology
artificial intelligence
colorectal cancer
diagnostic molecular pathology
personalised medicine

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1002/path.6376Licence: CC BY

Schoenpflug_etal_JoP_Tumour_Purity_Assessment_VOR
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. https://creativecommons.org/licenses/by/4.0/
Final published version, 5.78 MBLicence: CC BY

Cite this

Schoenpflug, L. A., Chatzipli, A., Sirinukunwattana, K., Richman, S., Blake, A., Robineau, J., Mertz, K. D., Verrill, C., Leedham, S. J., Hardy, C., Whalley, C., Redmond, K., Dunne, P., Walker, S., Beggs, A. D., McDermott, U., Murray, G. I., Samuel, L. M., Seymour, M., ... Koelzer, V. H. (2025). Tumour purity assessment with deep learning in colorectal cancer and impact on molecular analysis. The Journal of pathology, 265(2), 184-197. Advance online publication. https://doi.org/10.1002/path.6376

Schoenpflug, LA, Chatzipli, A, Sirinukunwattana, K, Richman, S, Blake, A, Robineau, J, Mertz, KD, Verrill, C, Leedham, SJ, Hardy, C, Whalley, C, Redmond, K, Dunne, P, Walker, S, Beggs, AD, McDermott, U, Murray, GI, Samuel, LM, Seymour, M, Tomlinson, I, Quirke, P, S:CORT consortium, Rittscher, J, Maughan, T, Domingo, E & Koelzer, VH 2025, 'Tumour purity assessment with deep learning in colorectal cancer and impact on molecular analysis', The Journal of pathology, vol. 265, no. 2, pp. 184-197. https://doi.org/10.1002/path.6376

@article{efb4d592717b44338dc95b0ccb145180,

title = "Tumour purity assessment with deep learning in colorectal cancer and impact on molecular analysis",

abstract = "Abstract Tumour content plays a pivotal role in directing the bioinformatic analysis of molecular profiles such as copy number variation (CNV). In clinical application, tumour purity estimation (TPE) is achieved either through visual pathological review [conventional pathology (CP)] or the deconvolution of molecular data. While CP provides a direct measurement, it demonstrates modest reproducibility and lacks standardisation. Conversely, deconvolution methods offer an indirect assessment with uncertain accuracy, underscoring the necessity for innovative approaches. SoftCTM is an open-source, multiorgan deep-learning (DL) model for the detection of tumour and non-tumour cells in H&E-stained slides, developed within the Overlapped Cell on Tissue Dataset for Histopathology (OCELOT) Challenge 2023. Here, using three large multicentre colorectal cancer (CRC) cohorts (N?=?1,097 patients) with digital pathology and multi-omic data, we compare the utility and accuracy of TPE with SoftCTM versus CP and bioinformatic deconvolution methods (RNA expression, DNA methylation) for downstream molecular analysis, including CNV profiling. SoftCTM showed technical repeatability when applied twice on the same slide (r?=?1.0) and excellent correlations in paired H&E slides (r?>?0.9). TPEs profiled by SoftCTM correlated highly with RNA expression (r?=?0.59) and DNA methylation (r?=?0.40), while TPEs by CP showed a lower correlation with RNA expression (r?=?0.41) and DNA methylation (r?=?0.29). We show that CP and deconvolution methods respectively underestimate and overestimate tumour content compared to SoftCTM, resulting in 6?13% differing CNV calls. In summary, TPE with SoftCTM enables reproducibility, automation, and standardisation at single-cell resolution. SoftCTM estimates (M?=?58.9%, SD ±16.3%) reconcile the overestimation by molecular data extrapolation (RNA expression: M?=?79.2%, SD ±10.5, DNA methylation: M?=?62.7%, SD ±11.8%) and underestimation by CP (M?=?35.9%, SD ±13.1%), providing a more reliable middle ground. A fully integrated computational pathology solution could therefore be used to improve downstream molecular analyses for research and clinics. ? 2024 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.",

keywords = "pathology, artificial intelligence, colorectal cancer, diagnostic molecular pathology, personalised medicine",

author = "Schoenpflug, {Lydia A} and Aikaterini Chatzipli and Korsuk Sirinukunwattana and Susan Richman and Andrew Blake and James Robineau and Mertz, {Kirsten D} and Clare Verrill and Leedham, {Simon J} and Claire Hardy and Celina Whalley and Keara Redmond and Philip Dunne and Steven Walker and Beggs, {Andrew D} and Ultan McDermott and Murray, {Graeme I} and Samuel, {Leslie M} and Matthew Seymour and Ian Tomlinson and Philip Quirke and {S:CORT consortium} and Jens Rittscher and Tim Maughan and Enric Domingo and Koelzer, {Viktor H}",

note = "The authors thank Claire Butler and Michael Youdell for excellent management in S:CORT and the MRC Clinical Trials Unit which provided the clinical data from the FOCUS trial with permission from the FOCUS trial steering group. We would further like to thank Indica Labs for providing the HALO{\texttrademark} software. The results published or shown here are based in part upon data generated by the TCGA Research Network established by the National Cancer Institute and National Human Genome Research Institute. Information about TCGA and the investigators and institutions who constitute the TCGA research network can be found at http://cancergenome.nih.gov. We would especially like to thank all patients who consented to take part in S:CORT and TCGA. The views expressed are those of the author(s) and not necessarily those of the National Health Service, the National Institute for Health and Care Research, or the Department of Health.",

year = "2024",

month = dec,

day = "22",

doi = "10.1002/path.6376",

language = "English",

volume = "265",

pages = "184--197",

journal = "The Journal of pathology",

issn = "0022-3417",

publisher = "Wiley",

number = "2",

}

TY - JOUR

T1 - Tumour purity assessment with deep learning in colorectal cancer and impact on molecular analysis

AU - Schoenpflug, Lydia A

AU - Chatzipli, Aikaterini

AU - Sirinukunwattana, Korsuk

AU - Richman, Susan

AU - Blake, Andrew

AU - Robineau, James

AU - Mertz, Kirsten D

AU - Verrill, Clare

AU - Leedham, Simon J

AU - Hardy, Claire

AU - Whalley, Celina

AU - Redmond, Keara

AU - Dunne, Philip

AU - Walker, Steven

AU - Beggs, Andrew D

AU - McDermott, Ultan

AU - Murray, Graeme I

AU - Samuel, Leslie M

AU - Seymour, Matthew

AU - Tomlinson, Ian

AU - Quirke, Philip

AU - S:CORT consortium

AU - Rittscher, Jens

AU - Maughan, Tim

AU - Domingo, Enric

AU - Koelzer, Viktor H

N1 - The authors thank Claire Butler and Michael Youdell for excellent management in S:CORT and the MRC Clinical Trials Unit which provided the clinical data from the FOCUS trial with permission from the FOCUS trial steering group. We would further like to thank Indica Labs for providing the HALO™ software. The results published or shown here are based in part upon data generated by the TCGA Research Network established by the National Cancer Institute and National Human Genome Research Institute. Information about TCGA and the investigators and institutions who constitute the TCGA research network can be found at http://cancergenome.nih.gov. We would especially like to thank all patients who consented to take part in S:CORT and TCGA. The views expressed are those of the author(s) and not necessarily those of the National Health Service, the National Institute for Health and Care Research, or the Department of Health.

PY - 2024/12/22

Y1 - 2024/12/22

N2 - Abstract Tumour content plays a pivotal role in directing the bioinformatic analysis of molecular profiles such as copy number variation (CNV). In clinical application, tumour purity estimation (TPE) is achieved either through visual pathological review [conventional pathology (CP)] or the deconvolution of molecular data. While CP provides a direct measurement, it demonstrates modest reproducibility and lacks standardisation. Conversely, deconvolution methods offer an indirect assessment with uncertain accuracy, underscoring the necessity for innovative approaches. SoftCTM is an open-source, multiorgan deep-learning (DL) model for the detection of tumour and non-tumour cells in H&E-stained slides, developed within the Overlapped Cell on Tissue Dataset for Histopathology (OCELOT) Challenge 2023. Here, using three large multicentre colorectal cancer (CRC) cohorts (N?=?1,097 patients) with digital pathology and multi-omic data, we compare the utility and accuracy of TPE with SoftCTM versus CP and bioinformatic deconvolution methods (RNA expression, DNA methylation) for downstream molecular analysis, including CNV profiling. SoftCTM showed technical repeatability when applied twice on the same slide (r?=?1.0) and excellent correlations in paired H&E slides (r?>?0.9). TPEs profiled by SoftCTM correlated highly with RNA expression (r?=?0.59) and DNA methylation (r?=?0.40), while TPEs by CP showed a lower correlation with RNA expression (r?=?0.41) and DNA methylation (r?=?0.29). We show that CP and deconvolution methods respectively underestimate and overestimate tumour content compared to SoftCTM, resulting in 6?13% differing CNV calls. In summary, TPE with SoftCTM enables reproducibility, automation, and standardisation at single-cell resolution. SoftCTM estimates (M?=?58.9%, SD ±16.3%) reconcile the overestimation by molecular data extrapolation (RNA expression: M?=?79.2%, SD ±10.5, DNA methylation: M?=?62.7%, SD ±11.8%) and underestimation by CP (M?=?35.9%, SD ±13.1%), providing a more reliable middle ground. A fully integrated computational pathology solution could therefore be used to improve downstream molecular analyses for research and clinics. ? 2024 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

AB - Abstract Tumour content plays a pivotal role in directing the bioinformatic analysis of molecular profiles such as copy number variation (CNV). In clinical application, tumour purity estimation (TPE) is achieved either through visual pathological review [conventional pathology (CP)] or the deconvolution of molecular data. While CP provides a direct measurement, it demonstrates modest reproducibility and lacks standardisation. Conversely, deconvolution methods offer an indirect assessment with uncertain accuracy, underscoring the necessity for innovative approaches. SoftCTM is an open-source, multiorgan deep-learning (DL) model for the detection of tumour and non-tumour cells in H&E-stained slides, developed within the Overlapped Cell on Tissue Dataset for Histopathology (OCELOT) Challenge 2023. Here, using three large multicentre colorectal cancer (CRC) cohorts (N?=?1,097 patients) with digital pathology and multi-omic data, we compare the utility and accuracy of TPE with SoftCTM versus CP and bioinformatic deconvolution methods (RNA expression, DNA methylation) for downstream molecular analysis, including CNV profiling. SoftCTM showed technical repeatability when applied twice on the same slide (r?=?1.0) and excellent correlations in paired H&E slides (r?>?0.9). TPEs profiled by SoftCTM correlated highly with RNA expression (r?=?0.59) and DNA methylation (r?=?0.40), while TPEs by CP showed a lower correlation with RNA expression (r?=?0.41) and DNA methylation (r?=?0.29). We show that CP and deconvolution methods respectively underestimate and overestimate tumour content compared to SoftCTM, resulting in 6?13% differing CNV calls. In summary, TPE with SoftCTM enables reproducibility, automation, and standardisation at single-cell resolution. SoftCTM estimates (M?=?58.9%, SD ±16.3%) reconcile the overestimation by molecular data extrapolation (RNA expression: M?=?79.2%, SD ±10.5, DNA methylation: M?=?62.7%, SD ±11.8%) and underestimation by CP (M?=?35.9%, SD ±13.1%), providing a more reliable middle ground. A fully integrated computational pathology solution could therefore be used to improve downstream molecular analyses for research and clinics. ? 2024 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

KW - pathology

KW - artificial intelligence

KW - colorectal cancer

KW - diagnostic molecular pathology

KW - personalised medicine

U2 - 10.1002/path.6376

DO - 10.1002/path.6376

M3 - Article

SN - 0022-3417

VL - 265

SP - 184

EP - 197

JO - The Journal of pathology

JF - The Journal of pathology

IS - 2

ER -

Tumour purity assessment with deep learning in colorectal cancer and impact on molecular analysis

Abstract

Bibliographical note

Data Availability Statement

Keywords

UN SDGs

Access to Document

Fingerprint

Cite this