Machine learning prediction of genetic mutations impact

Overview

Cancer is caused by a sequence of acquired somatic genomic aberrations. Subset of tumors are familial and occur on a background of a germline mutation. Recent advances in cancer genome sequencing led to large-scale projects characterizing the genomic landscape of many tumor types. These projects contributed to the identification of cancer genes involved in specific tumor types and in addition, it was realized that each patient has a unique mutations profile. Recent developments in sequencing technology made tumor genomic analysis feasible for individual patients. In parallel, many new drugs, targeting specific cancer genes have been developed and approved.

These two coupled endeavors enabled personalized medicine approach by which treatment is tailored for the individual patient based on the tumor genomic profile. This approach already showed impressive efficacy for subset of cancer patients and thus hold promise for wider application of personalized medicine approach in cancer treatment.

Despite these advances most patients do not benefit from precision oncology approach. One of the major challenges is to understand the biological importance of every genetic mutation identified in a cancer patient. This will enable to prioritize personalized treatment for the relevant somatic mutations. In addition, it will enable to provide better genetic consultation for carriers of germline mutations.  

The Laboratory for Computational Biology of Cancer aims to harness the power of big data and artificial intelligence to investigate and address these unmet needs.

Our Solution

Many mutations in cancer genes are defined as Variants of Uncertain Significance (VUS). It is crucial to discern pathogenic from the neutral variants. This is important for precision oncology as well as for genetic consultation for carriers of such variants. The Hadassah researchers created TP53_PROF, a gene-specific machine learning model to predict pathogenicity for all possible missense mutations in TP53 gene. The unique gene-specific model surpassed state-of-the-art tools and reached accuracy of 96.5%. Such outstanding result was extensively validated using experimental and clinical data (Ben-Cohen et al., 2022). This approach can be applied and generalized to other cancer genes.

• The model can predict every missense variant, including those that did not occur yet

Applications

• Germline variants: genetic consultation for carrier of mutations
• Somatic variants: for precision medicine in cancer – identification of mutations to target

IP

An international patent application was filed (PCT)

References

Ben-Cohen, G., Doffe, F., Devir, M., Leroy, B., Soussi, T., & Rosenberg, S. (2022). TP53_PROF: a machine learning model to predict impact of missense mutations in TP53 . Briefings in Bioinformatics, 23(2), 1–19. https://doi.org/10.1093/bib/bbab524