Ghent, Belgium, 4 November 2020 – myNEO, a developer of personalised therapeutics using the patient’s tumour-specific alterations, today announced the launch of the next generation of its ImmunoEngine bioinformatics platform. The update includes two, one-of-a-kind machine learning algorithms, neoMS and neoIM, which enable the accurate and sensitive analyses of tumours. Specifically, the ImmunoEngine technology searches for the unique neoantigens found on the surface of cells that are different between tumour and healthy cells in each cancer patient. This is a critical first step in the development of a therapeutic vaccine, which uses the patient's immune system to eliminate tumour cells. myNEO believes this more personalised approach could be used to help treat the large number of cancer patients who do not benefit from existing, more general immunotherapies, particularly those patients with difficult-to-treat tumours that currently have very limited treatment options.
During immunotherapy, the immune system is activated in the vicinity of the tumour so that it can eliminate the cancer cells. Using a tailor-made solution, in which the body is helped to clear the tumour cells on its own, is a much more elegant approach than alternatives, such as chemotherapy, where a toxic substance is injected that harms all dividing cells, including those that are healthy. - Cedric Bogaert, CEO of myNEO.
Every person develops abnormal cells every day due to errors during the reading of the DNA (mutations), but they only rarely grow into real tumours. The body's immune system is not only responsible for eliminating viruses and bacteria, but also for the detection and removal of these abnormal cells. If the immune system does not respond to these cells, it can cause cancer.
Some tumours differ little from normal healthy cells, so they are often not recognised as “foreign” by an active immune system. By specifically providing the immune system the differences between the tumour and the healthy cells for a particular patient, in the form of a personalised treatment, myNEO believes its approach will promote the removal of these tumours. This individualisation of treatment provides the immune system with the information it needs to correctly distinguish tumour cells from a healthy cell and develop a strong and sustainable anti-tumour immune response.
The myNEO ImmunoEngine technology makes it possible to detect foreign proteins on the tumour for each patient even if the tumour differs very little from normal cells. This is because myNEO performs a broader genomic screening of the tumour compared to other technologies, which increases the chance of finding these proteins. myNEO has developed and incorporated two new machine learning algorithms, neoMS and neoIM, which together allow for very sensitive and accurate prediction and selection of these hard-to-find, foreign proteins.
The next generation myNEO ImmunoEngine technology, improved with the patented neoMS (presentation prediction) and neoIM algorithms (immunogenicity prediction), paves the way for better neoantigen prioritisation and selection enabling the development of more potent vaccines. The algorithms have shown to outperform the current industry standards revealing the promise, uniqueness and strength of the myNEO technology. It is myNEO’s goal to bring its technology to the clinic focusing on patients with difficult-to-treat tumours that often relapse and have no effective treatment options. - Bruno Fant, CTO of myNEO.
The neoMS algorithm
The neoMS algorithm is one-of-a-kind and is a transformer-based, patient-oriented presentation prediction algorithm trained with MHC ligandomic data. This in contrast to the industry standards, which still use MHC binding as a proxy for presentation, fail to take all other steps of the presentation machinery into account and are unfit to assess the potential clinical benefit of a given epitope.
The neoMS presentation prediction algorithm achieves robust predictive power, given as input a peptide and an HLA allelic set, and is able to reliably predict MHC-I presentation at the surface of cancer cells. The model achieves up to 0.75 precision at recall 0.8 on its test set, vastly outperforming the current industry standard netMHCpan (with a max precision of 0.23 across any allele of the test set considered). In addition, due to his sequence-based comparison method, neoMS exhibits extrapolation capabilities, achieving non-zero predictive power when evaluated on ground truth ligandome data derived from an HLA allele completely absent from the training set.
The neoIM algorithm
The neoIM algorithm is based on the necessity of identifying only those peptides that can elicit an immune response for personalised cancer vaccine development. Current immunogenicity assays are expensive and time-consuming. In addition, research on immunogenicity prediction, i.e. predicting the likelihood of T-cells to recognise and react to a peptide presented on MHC-I, is still lacking behind.
The neoIM algorithm is a first-in-class, random forest classifier specifically trained to classify short peptides of length 9-11 amino acids as immunogenic or non-immunogenic. The model achieves prediction of peptide immunogenicity with high accuracy (AUC=0.84), outperforming the currently available methods INeo-Epp (AUC=0.65) and IEDB Immunogenicity Predictor (AUC=0.57).