Kanika Leang, a junior in biochemistry at the University of Illinois at Urbana-Champaign, used machine learning to predict gene dependencies from mutation data.
Gene-dependent cancers are driven by the effects of one or a few genes. Since each cancer is so unique, it’s important that cancer treatments are also specific to the patient. When first joining the Broad, I was motivated by the research opportunities available to me as a scientist. However, it wasn’t until the program started that I truly understood what this experience offered me. Through leadership courses and communication seminars, I was inspired by the inclusive and collaborative efforts at the Broad. I can’t wait to use my new skills in computational genomics, and as a leader, at my home institution and beyond!By targeting the genes which contribute the most to cancer growth, we can administer more personalized and overall effective treatments to each patient’s cancer. Genes have the potential to either support or suppress cancer growth. This potential can be quantified into a gene dependency score, which is measured by how well a cell grows or dies without a given gene. Gene dependency can be instrumental in determining a given gene’s contribution to cancer growth. Though, gene dependencies can’t be retrieved directly from patient data. Instead, we used gene features such as whether a gene is mutated in a cancer cell, to predict a given gene’s dependency score. Using the computer program, R, we designed a linear model. The accuracy of this model was assessed using the oncogene KRAS. The model showed some predictive power for KRAS with a high correlation of about 0.75. In the future we plan to assess this model on other well-studied oncogenes. This knowledge will be useful in prioritizing the greatest cancer-contributing genes to target with cancer therapies.
Project: Predicting Gene Dependencies Using Mutation Status
Mentors: Kasia Handing, Center for the Development of Therapeutics (CDoT)
David Wu, Cancer Data Science Project