Owkin

Published by Daniel at February 1, 2022

Owkin is a research network co-founded by Thomas Clozel, MD, a former assistant professor in clinical hematology/oncology, and by Gilles Wainrib, Ph.D., and former researcher about machine learning systems and their application in biology and medicine. The company was established assuming that medical research should be based on collaboration, inclusion, and privacy. Owkin is building a network that connects researchers, scientists, medical personnel, and drug companies keeping the information secure and private. Owkin’s work has been published in multiple scientific journals, including Nature Medicine. It has raised over $70 million in financing and is now working with the most important cancer centers and pharmaceutical companies in Europe and the US.

Owkin AI Principles

Owkin aims to give researchers in hospitals, universities, and pharmaceutical companies all the necessary tools to achieve their objectives.

Patients have a desire for faster development of safer and more practical therapies.

Owkin gives the necessary tools to researchers from hospitals, universities, and pharmaceutical companies to:

• Establish why drugs affect some patients differently.

• Enhance the drug development process.

• Identify the most effective drug for the correct patient to improve treatment outcomes.

Therefore, Owkin created a novel research platform and a portfolio of AI models and solutions called Owkin Loop. It is defined as the research basis

of Owkin’s platform: it connects healthcare researchers with databases from multiple academic research servers worldwide.

Owkin Loop is powered by the two main components of Owkin’s Software Stack:

• Owkin Studio, their ML platform.
• Owkin Connect, their federated learning framework

Owkin’s medical research collaborations include oncology, immunology, and cardiovascular diseases.(1)

Owkin IT Aspects

Owkin lab: the team is a combination of multiple ML scientists, data analysts, engineers, biologists, medical doctors, biostatisticians, and translational researchers. The Lab is divided into teams dedicated to a specific data modality, including clinical data, radiology, pathology, and genomics.

Owkin Loop: This software allows teams access to high-quality patient data and top-notch clinical experience through globally connected networks of academic medical centers. Owkin aims to the highest quality, research-grade cohorts that contain longitudinal molecular, imaging, pathology, and outcome data. Owkin Connect powers Owkin Loop.

Owkin Connect: it is the framework where the Federated Learning system operates across Owkin Loop. It provides information to train AI models on decentralized data, which solves the data sharing challenges in healthcare. The data is kept secure and private, on-site behind the medical center’s firewall. Owkin Connect empowers data owners to determine data authorizations and track their usage.

Federated Learning: It is an ML system with the objective of training AI models from multiple independent providers.

Instead of keeping data on a single server, the data remains locked on their server, and the algorithms and predictive models move between them. The main goal of this approach is to take advantage of a large pool of data resulting in increased ML performance while respecting data ownership and privacy.

Owkin Studio: it is a platform designed to articulate the capabilities of AI and the experience of clinical researchers. The user structures a research project that trains an ML model and interprets the result. Owkin’s studio is engineered to allow non-ML experts to perform AI research projects. The software identifies the features in the data responsible for predictions and allows the researchers to investigate their biological meaning.
(2)

Owkin as an Investment

Last year, Owkin raised $25 million in financing from multiple investors, including Bpifrance, Large Venture, Cathay Innovation, and MACSF, including previous investors such as GV, F-Prime Capital, and Eight Roads. The $55 million total capital investment from established and new investors comes from the company’s extended Series A round. In 2020, the company started The Covid-19 Open AI consortium, an initiative that employs their platform to maintain and develop collaborative research. The company’s estimated annual revenue is 20.3 million per year and grew its employee count by 24% compared to last year.(3)

Development of AI-based Pathology Biomarkers in Gastrointestinal and Liver Cancer

Studies show that numerous human malignancies can be identified from histopathological images, including gastric and colorectal cancer. With the digitization of histopathology slides in clinical routine, deep learning (DL) systems could be trained on histological images and use the Response Evaluation Criteria in Solid Tumours (RECIST) as the labels to be assigned to the samples.(4)

Deep Learning Model Can Predict RNA-Seq Expression of Tumors

HE2RNA, a DL algorithm, can be trained to predict genes involved in immune cell activation status and immune cell signaling; therefore, it could predict immune infiltration and participate in oncology treatment decisions in the context of immunotherapy. HE2RNA can also predict the genes involved in immune regulation and track different types of cancer.(5)

Abdominal Musculature Segmentation CT Using Deep Learning for Sarcopenia Assessment

The research group built a Convolutional neural network (CNN) system to segment muscular body mass to predict muscular surface from a two-dimensional axial CT scan slice through the L3 vertebra. A total of 1025 individual CT slices were used for training the system, and 500 were used for testing. The results showed that the system could provide new biomarkers for a better diagnosis of sarcopenia.(6)

Integration of Clinic, Laboratory Tests, and CT scan to Predict Severity of Hospitalized COVID-19 Patients

Early detection of high-risk COVID-19 patients during admission and hospitalization is primordial to deliver proper medical attention and optimize the use of medical areas, especially the already limited intensive care units. A French research group designed a study integrating clinical, biological, and radiological data to predict the outcome of hospitalized patients. They collected 58 clinical and biological variables, chest CT scan data (506,341 images), and radiology reports from 1,003 coronavirus-infected patients at hospital admission. With these data, they developed a DL system (AI-severity) that included five clinical and biological variables (age, sex, oxygenation, urea, and platelets) as well as a CT DL model.(7)

The results showed that the 6-variable AI-severity score integrated with a radiological quantification of lesions with key clinical and biological variables provided accurate severity predictions compared with the other 11 existing severity scores. Indicating that the AI-severity can rapidly become a reference tool for evaluating hospitalized patients.(7)

Owkin IT Aspects

Multiparametric models may predict cancer evolution in the future, depending not only on tumor features but also on peri-tumoral environmental and individual-derived information. Patient anthropometrics can be estimated using variables such as weight, height, or waist circumference. CT scans and MR imaging allow to measure the adipose or muscle compartments, and provide 3D, whole body, and high-resolution images of body composition.

The purpose of the study is to calculate accurate measures from CT with AI and ultimately assess the prognostic value in patients with non-small-cell lung cancer (NSCLC). The group trained a convolutional neural network to automatically segment subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), and muscular body mass (MBM) from the low-dose CT scans of 189 patients with NSCLS who underwent pre-therapy PET/CT scan. The results indicated that the Body surface area-normalized VAT/SAT ratio is an independent predictor of progression-free survival and overall survival in NSCLC patients.(8)

References