Sanofi, MIT Jameel Clinic join to advance health research with AI
The Abdul Latif Jameel Clinic for Machine Learning in Health at the Massachusetts Institute of Technology (MIT) is forging a long-term partnership with Sanofi. The research teams will work together to develop and apply artificial (AI) and machine learning (ML) technology to advance drug discovery, development of treatments and other projects.
Outsourcing-Pharma (OSP) recently discussed the partnership with Ignacio Fuentes (IF), managing director at the Jameel Clinic, as well as how advanced technologies like ML and AI are transforming the life-sciences landscape.
OSP: Could you please provide an overview of how machine learning and its various applications has evolved in recent years?
IF: Machine learning has evolved at a blistering pace over the past few years, accelerated by smarter model architectures, larger datasets, and larger compute capabilities.
On the natural language processing side, models have gained a stronger ability to mimic human language by generating increasingly natural sentences, as well as a better understanding of language. On the computer vision side of machine learning, there have been many advancements in image segmentation, where models can pick out all the different objects in a natural scene.
Smarter architectures have also allowed models to decide what parts of an image or video it should attend to.
More recently, there has been a strong push to develop models that require very little supervision, where we no longer have to tell models explicitly whether something is this or that. In a similar vein, there have also been advances in developing models that can learn from few examples, as opposed to requiring millions of examples in order to learn and achieve high performance.
Inspired by how humans learn their behavior, reinforcement learning has also enjoyed success in training models on increasingly complex tasks (from dynamically modulating the temperature of a warehouse to playing Atari games). In fact, the list of developments goes on — whether it’s to generate realistically-looking images, design secure and private models, or remove acquired biases from models’ predictions. There has been a flurry of developments over the past few years.
OSP: Specifically, how has ML been applied in healthcare research and innovation?
IF: In healthcare, ML has been primarily applied to areas where the data is rich and available. Since some cancer screens are standardized (mammography is the quintessential example), there have been deep learning models developed for detecting breast cancer and lung cancers. Models have also been built to diagnose respiratory illnesses from chest X-rays, classify cardiac arrhythmias from EKGs, and predict malignant tissues from histopathology slides.
There has also been work on the basic side of healthcare. For instance, models are getting increasingly better at predicting protein folding, discovering new drugs, and characterizing the function of genes. The innovation in this space is certainly just beginning.
OSP: How did the Jameel Clinic come to partner with Sanofi? Have your two organizations collaborated in the past?
IF: MIT has collaborated with Sanofi in the past but this is the first time that Jameel Clinic, an MIT machine learning initiative in health care, has established a collaboration with Sanofi. It was a matter of timing and opportunity, with Jameel Clinic expanding its research work in therapeutics AI and Sanofi looking for an opportunity to be a more active player in the MIT Boston pharma and biotech ecosystem.
OSP: Is there anything you’d like to add about the collaboration, the clinic, your other work, etc?
IF: This is a great opportunity for a partnership of this nature. Jameel Clinic has been very active in clinical and therapeutics AI, developing state of the art models that now, thanks to this collaboration, can be trained using unique proprietary datasets. A number of joint milestones including education programs and access to the MIT research community will enhance this collaboration over time.