Immunotherapy response prediction

Context

A few years ago, the pharmaceutical company GSK developed immuno-therapies against several forms of cancer. One of them, was a therapeutic vaccine targeting melanoma. Within a Phase II trial, several melanoma patients of a cohort were responding to the treatment while other were not, lowering the global efficacy of the treatment. Transcriptomic data had been taken for several tens of thousands of potential markers before treatment was given to the patients.

Objective

We were asked if our technology would be able to predict the response to the treatment based on those data, and at the same to time identify which smallest possible set of markers was needed to make such prediction. Those objectives were the basis for patient stratification that had to be implemented in a Phase III trial. Patient stratification would lead to increased efficacy on the targeted profiles. Responder profiling would also have allowed for the development of companion diagnostics strategies.

What we did

We received the data and the data had been blinded. Not only had the names of the patients been removed (which is always the case) but the names of the 54,000 candidate markers had also been removed. We applied a robust method for biomarker signature identification and returned a set of 33 variables, in the form of a series of indexes (“our algorithm picked feature 2, feature 1294, feature 3141,…” and so on). We also built and validated a model using this gene signature for predicting the treatment response or not prior to treatment administration. Some of the results from this collaboration gave rise to a patent application on a method for classifying a cancer patient as responder or non-responder to immunotherapy.

Customer’s feedback

“DNAlytics technology has been used to seek predictive markers associated with clinical benefit to our cancer immunotherapeutic in melanoma patient. We provided them blinded data for more than 54.000 expression markers, but without the name of the markers and binary information on clinical outcome. After few weeks, they provided a list of 33 markers that were discriminative for patient outcome. Unblinding of the code showed that the gene signature was biologically very relevant, and confirmed biological hypotheses stated by our experts! The data were patented and are being prospectively validated in a PhII metastatic melanoma study treated with our lead cancer immunotherapeutic.”
Jamila Louahed,
Vice President, DAP Head R&D ASCI,
Immunotherapeutics Business Unit GSK Vaccines.

What if it had to be re-done?

Unlike in many of our more recent projects, we did not receive much clinical data for this cohort (apart from the outcome). Had we received more clinical data, we would have tested solutions relying on hybrid signatures, mixing transcriptomic and clinical information, which often proves more powerful.

Do you have to deal with a similar project?

Contact us!