Deep learning model to identify biological processes active in tumors

Overview

Cancer is the leading cause of death in the western world. Previous medical practice was to treat all patients with a given tumor type in a one-size-fits-all manner. However, tumors are caused by unique mixture of mutations in many cancer genes. In the last two decades we learnt that every tumor has its own fingerprint of disrupted genes. Many drugs were developed to target these genes and this paved the way to personalized treatment of each patient according to his/her mutations fingerprint. This approach led to tremendous success in subtypes of lung cancer and melanoma, among others. However, only minority of cancer patients derive actual clinical benefit from this approach, let alone actual curation. Several reasons for this are:

• Lack of methods to predict the patient’s chance to respond to the prescribed precision treatment. Hence, the optimal treatment is difficult to define.
• Many important treatable genomic changes are not properly identified and hence treatments with good potential are not even considered.
• Mostly, mutations of the tumor are evaluated as an isolated event and not systematically as part of the complete genomic landscape of the tumor. Consideration of the complete tumors genomic fingerprint might be necessary to prescribe the optimal treatment.

Our solution

• Deep learning model to identify biological processed active in tumors.
• This method includes ~300 such models that are specific to tumor types and biological processed (as more data becomes available more models will be developed)
• The models can quantify the relative strength of the activity of the biological processed in given tumors

Applications

• Drug development
• Identify clinically meaningful targets in given tumors
• Prioritize biological processes and genetic mutations for drug targeting

IP

Priority application filed