How an AI tool predicts responses to cancer treatment
Original source here
Ofer Sharon, MD, CEO of OncoHost, spoke to Med-Tech Innovation News about its offering in oncology – and how its AI tool predicts responses to treatment.
Tell us about your company. When did you establish yourself and what do you offer?
OncoHost, founded in 2017, is a clinical stage precision oncology start-up that combines proteomic analysis with AI to predict response to immunotherapy and provide clinicians with potential combination strategies to overcome treatment resistance.
Our ultimate goal is to develop a comprehensive platform that will be able to predict response to immunotherapy and provide clinicians with a “cancer management GPS” that will support oncologists as they navigate the complex clinical decisions of cancer treatment.
OncoHost has developed a first of its kind host response profiling platform (PROphet) for predicting responsiveness to cancer therapies. Our technology analyses proteomic changes in patient blood samples to detect signs of resistance to cancer therapies in real-time, enabling biomarker-guided treatment planning for physicians, target discovery for drug development and, ultimately, improved outcomes for patients.
We are currently focused on melanoma skin cancer and small-cell lung cancer and will soon expand to other types of cancer. So far, our system is showing about 90% accuracy in predicting how melanoma and lung cancer patients respond to various therapies. We plan to launch our product in the market in 2021.
Where did the idea for your start-up come from?
The idea behind OncoHost began with a burning question: Why do cancer treatments help some patients but not others?
Our co-founder and chief scientist, Prof. Yuval Shaked, head of the integrated cancer research centre at the Technion – Israel Institute of Technology, discovered that a patient’s biological response to anti-cancer treatment, known as the ‘host response,’ may actually facilitate and support tumour growth and spread. In what seems like a paradoxical reaction to treatment, our own body may actually help the tumour cells evade the effect of the cancer treatment.
Over 15 years of research, Shaked discovered that host response is a universal phenomenon regardless of the treatment modality (chemotherapy, targeted therapies, immunotherapy and even radiation and surgery). This encouraged us at OncoHost to develop a system that can decipher the host response in order to identify which treatments are best suited for individual patients, and which should be avoided.
Our system, PROphet, can identify up to 1,000 possible proteins in a patient’s blood plasma. It then uses machine learning to comb through thousands of patient blood profiles, identifying how different patterns of proteins were associated with success in various immunotherapy treatments and flagging the therapies most likely to help an individual patient.
What difference do you think you can make in your particular sector?
Cancer care made significant progress in the last decade with the introduction of modern immunotherapies. We were hoping for a ‘silver bullet’ with a drug mechanism that could potentially be universal, with what looked like a favourable adverse event profile. Today we know that the challenge is still ahead of us, as most patients do not respond to immunotherapy, and while adverse events are different, they are indeed still there. Immunotherapy doesn’t work for all patients, and it can often take a few months to evaluate if the treatment is working or not. We are still lacking good biomarkers that can tell us which patients will respond and which will not.
Most precision oncology platforms are focused on the interaction between the cancer therapy and the tumour. This approach makes a lot of sense when we’re looking for new treatment targets but, unfortunately, it makes less sense when we’re looking for biomarkers. The reason for this is that the therapy-tumour interaction takes place within a complex biological system – the human body. We know today that the body’s response to treatment (AKA host response) significantly effects the therapy-tumour interaction. Our science is focused on deciphering the therapy-tumour-host circle.
Another important differentiator for OncoHost is our decision to focus on proteomics. Proteins represent, and are a part of, biological process. When we find a certain protein, it means it had a role in the biology of the patient. In contrast, most precision oncology companies are focused on mutations, genetic alterations in the tumour cells that may have a role in the development and resistance of the tumour (e.g., the driver mutations) or not have a role at all. We know that in many cases, an alteration in the DNA level doesn’t always translate to a biological process.
Our technology, PROphet, will be first made available to healthcare providers who can use it to help assess how a specific patient may respond to a given treatment. This will aid the oncologist and patient in making informed decisions regarding treatment options, lines of therapy, potential combination treatments, and participation in clinical trials.
Tell us more about the technology at the centre of your product?
PROphet is a first-of-its-kind system that analyses proteomic changes in blood samples to monitor the dynamics of biological processes induced by the patient (i.e., the host) in response to a given cancer therapy. This proteomic profile is highly predictive of individual patient outcome, thus enabling personalised treatment planning. PROphet also identifies potential drug targets, advancing the development of novel therapeutic strategies.
Our lab uses high-throughput protein analysis technology to quantify the levels of thousands of proteins in a single plasma sample. The proteins include cytokines, chemokines, growth factors and enzymes associated with therapy resistance and tumour spread. We have optimised the system in terms of quality control, technical handling of samples, running multiple assays simultaneously, and selection of factors to be analysed.
PROphet employs powerful bioinformatic tools and machine-learning algorithms to identify host response proteomic patterns associated with resistance to therapy. By analysing proteomic changes in patient blood samples, our system monitors the patient’s response to treatment in real-time, and accurately predicts treatment outcome.