Today’s patients are tomorrow’s therapies — Why we invest in Owkin
We have recently announced leading a 11m$ investment in Owkin, a startup that applies transfer and federated learning to medical data to accelerate the discovery and development of new therapies.
On the back of several partnerships with leading hospitals (Institut Curie, Centre Léon Bérard) and pharmaceutical companies (Amgen, Actelion…) and credited with a 10th position in Kaggle’s Data Science Bowl 2017, Owkin has positioned itself among the leading innovative startups in the AI for life sciences space in less than a year since its launch.
Pharma’s traditional business model is not profitable anymore
Historically, the pharmaceutical industry’s business model has relied on intense R&D activity (~160Bn$ spend / year, 22% of net sales) in order to ensure a healthy stream of future revenues.
This model has been challenged for some years now by several factors : increased difficulty to identify actual breakthroughs, strong growth of generic and bio-similar drugs, strong pressure on price by regulatory bodies and public opinion to name but a few.
As a result, ROI for new molecules came tumbling down :
A new molecule costs 3Bn$ of R&D and takes 13 years to develop on average — a lengthy and still highly iterative and manual process. This high cost also hinders the industry’s ability to explore new areas (orphan drugs, targeted therapies, personalized medicine,…) where the number of target patients is too small to ensure a sufficient market size — the economics just don’t work.
The improvement of R&D productivity is a key challenge for the industry — and a multi-faceted one with several areas of improvement :
- Prioritize higher outcome molecules (i.e. with a higher benefit/risk ratio for patients & society as a whole)
- Increase clinical trials’ success rate
- Optimise R&D costs (medical costs as much as related G&A expenses — CROs notably)
- Accelerate the development cycle
A massive trove of data — but behind multiple fences and well-guarded gates
Machine learning, which is all the rage now, and in particular NLP and deep learning, is seen as one of the obvious candidate to bring this much-needed productivity to many steps of the R&D chain — making the process more data-driven, allowing to simulate and predict in silico how a molecule will behave, and helping better targeting of patients & selecting centers ; and many of this before even entering in the lengthy — and costly — process of running a clinical trial.
This is all well and good. Unfortunately this also runs into a massive roadblock : the utter sensitivity of all the stakeholders to data protection, be it for privacy reasons, trade secret reasons or simply regulatory reasons. The recent uproar about Deepmind’s deal with the NHS in the UK is a testament to this sensitivity.
If properly managed though, this treasure trove of data could help better categorise patients to do more accurate diagnoses, identify new therapies and monitor drugs in the post-market (Docétaxel anyone ?)…
The algorithms pass, the data stays
Thomas & Gilles, Owkin’s cofounders, have built a world-class team of AI researchers & engineers in no time, in order to bring about an innovative and ethical solution to this problem.
Enters Socrates — an AI platform for discovery, allowing researchers in hospitals and laboratories to leverage a library of pre-trained algorithms in order to identify exclusive insights in their proprietary data (medical imaging, blood tests, patient data, doctor’s diagnoses, genomics, epigenetics, …). The goal is to find new biomarkers that can help diagnose a patient, predict an illness’ activity/strength or the relapse risks, or predict a patient’s potential response to a new drug.
Socrates’ secret weapon is that it leverages transfer learning and federated learning, cutting-edge machine learning techniques. Transfer learning allows an algorithm to learn faster on a given dataset thanks to its pre-training on another dataset ; and federated learning allows to train a model on distributed data without needing to pooling it.
This allows Socrates to capitalize on the collective knowledge built up in other areas in order to provide extremely accurate analyses, even on small datasets.
Even more importantly, this also allows Socrates to learn across the datasets of all the users of the platform, without their data ever leaving their premises or being shown to anyone. This ensures the utmost level of privacy and security to the users, fully adapted to the very high standards required in the industry.
Tbe ability to leverage the massive amount of patient data available in the real world while abiding by very stringent confidentiality and security rules is the key to unlock a whole new area of productivity for the R&D process of the pharma industry and enable the development of new, targeted therapies.
This is the mission of Owkin’s stellar team of AI researchers, engineers and doctors.
Looking further down the road, their approach will also enable the rise of innovative models to share the value extracted from the data among the various participating hospitals, based on their datasets’ relative contribution to the algorithm’s improvement ; thereby providing them with new revenue streams that could contribute to lowering the healthcare bill footed by the patient and the taxpayer.
Sources
(1) Deloitte study, 2016
(2) Paul, Mytelka et al,How to improve R&D productivity:the pharmaceutical industry’s grand challenge — Nature Mars 2010; PwC study — Pharma 2020 Marketing the future