Owkin publishes breakthrough research in Nature Medicine
Owkin, which is developing machine learning technologies to advance medical research, today announces publication of a paper in Nature Medicine that showcases its breakthrough analysis of tumour biology using interpretable deep-learning models.
The paper entitled ‘Deep learning-based classification of mesothelioma improves prediction of patient outcomes’ describes how Owkin has developed a detailed and accurate prognostic model based on images of lung tissue biopsies to predict disease evolution and to identify associated biological features in mesothelioma.
Thomas Clozel, Chief Executive Officer, Owkin commented: “Mesothelioma is an aggressive cancer that often attacks the lining of the lungs and is frequently associated with asbestos exposure. Sadly, it proves to be fatal for most patients. Patients with mesothelioma exhibit a very high variability in survival, from a few months to a few years, and this makes it challenging for doctors to plan treatment and to care for these patients. Our research helps to explain the biological causes of this variation and will ultimately lead to the development of more targeted drugs and better management of this terrible disease.”
Owkin’s deep learning model, called MesoNet, was used to analyse digital Whole Slide Images (WSI) of pleural surgical biopsies from nearly 3000 mesothelioma patients.
To train and test MesoNet, Owkin used the extensive dataset from MESOBANK, which sources its data from multiple French institutions. Medical experts in pathology at the renowned French Cancer Institute, Centre Leon Berard (CLB) provided expert validation of the model results, confirming that MesoNet outperformed all existing survival models and demonstrated robustness to heterogeneity when it was successfully validated on differently stained images from The Cancer Genome Atlas (TGCA).
Beyond its predictive performance and this novel way to characterize mesothelioma subgroups, the deep learning model developed by Owkin was also able to highlight precise regions of interest in the image that are associated with the prognosis prediction. This key interpretability feature, combined with an original iterative collaboration with expert pathologists using Owkin software platform, has led to the identification of novel biological features that supports a deeper explanation of heterogeneity in this disease.
Françoise Galateau, Professor of Pathology at Centre Leon Berard, commented: “It was a great experience for our lab to work closely with the Owkin scientists to identify new subgroups within our patient population. The collaboration exceeded our expectations. As well as improving our prognostic models, MesoNet was able to identify new biomarkers within the stromal regions of the tumour microenvironment that were predictive of survival. This ability really sets Owkin’s AI models apart and has given us new direction in our research into mesothelioma.”
Owkin is now working with its partners in the biopharmaceutical industry to use this insight for enriching patient selection in clinical trials and identifying which patients are most at risk and therefore best suited for new druggable approaches in a trial setting.
Owkin’s innovative model of collaboration between academia and the biopharmaceutical industry is generating new insights from real world evidence captured from patients in trials and clinical practice.
It has set up Owkin LOOP, a federated network of US and European academic medical centres, which includes Centre Leon Berard.
Gilles Wainrib, Chief Scientific Officer, Owkin, added: “Owkin’s Technology enables our algorithms to learn from the patient data within hospital firewalls, without removing the data from the hospital. This scalable approach protects patient privacy and assures our hospital partners and their patients that the data is kept safe and secure. This capability, alongside the interpretable approach to AI discussed in this Nature Medicine paper (NMED-L95980B), is fast making Owkin LOOP the leading destination medical researchers and drug development professionals to gain predictive insight and leverage nextgeneration biomarkers. Ultimately, we hope that it will accelerate development of better therapies for patients with the greatest need.”