Fellow’s research: Employing Artificial Intelligence for Genomics-Based Prognosis of Patients with Blood Cancer

18 Oct 2019

Fellow’s research: Employing Artificial Intelligence for Genomics-Based Prognosis of Patients with Blood Cancer


Dr Nikhil Patkar, Intermediate Fellow

Tata Memorial Centre, Mumbai, India

Our recently published study, demonstrates the applicability of artificial intelligence (AI) towards implementation of personalized medicine. In a proof of concept, we demonstrate that machine learning, which is a subtype of AI, can be used for cancer risk prediction using complex genomics datasets.

Cancer is caused by the genetic changes that are acquired by our cells as we age. In the last few years, cutting-edge DNA sequencing methods have been put to use to identify these genetic changes or mutations, and we have learned much about the cancer genome. However, we are struggling with relatively simple questions: How do we know which mutations are important for a patient when multiple mutations co-occur? How do we make sense of genomic data? Until now researchers have resorted to “cherry picking” mutations and extrapolating their clinical relevance. However, this approach is biased as it is limited by the human inability to see existing connections in a maze of complex data.

Our study focussed on a cancer of the blood-forming cells called acute myeloid leukaemia (AML). AML with mutated NPM1 gene (NPM1mut AML) is one of the commonest subtype of AML. In our dataset, genome-sequencing data identified close to 400 mutations among 110 NPM1mutAML patients (≥18 years). Using our AI algorithm (called supervised machine learning), we identified features most likely to influence the disease outcome and developed a numerical score for every feature. A final sum of the scores enabled classification of this AML into three classes based on the likely course of the disease.

We have previously demonstrated the utility of immunophenotyping-based measurable residual disease (MRD), which is a measure of low level leukemic clones not visible by microscopy (1). Presence of MRD inevitably heralds the onset of relapse much before it occurs and thus AML MRD is important to predict outcome and guide further treatment strategies. We showed here that the three AI determined prognostic risk classes correlated with immunophenotyping based MRD. Furthermore, we demonstrate that these AI determined classes showed a high correlation with treatment outcome. Patients who were in the poor genetic risk category had a much shorter survival and were eight times more likely to relapse as compared to patients in the favourable genetic risk category. Even though our study includes a relatively small cohort and retrospective analysis, we believe findings of this study may pave the way for integration of AI in cancer genomic medicine.


1. Utility of Immunophenotypic Measurable Residual Disease in Adult Acute Myeloid Leukemia –Real World Context. Nikhil Patkar, Chinmayee Kakirde, Prasanna Bhanshe et al. Frontiers in Oncology. June 2019.

2. A Novel Machine Learning Derived Genetic Score Correlates with Measurable Residual Disease and is Highly Predictive of Outcome in Acute Myeloid Leukemia with Mutated NPM1. Nikhil Patkar, Anam Fatima Shaikh, Chinmayee Kakirde et al.  Blood Cancer Journal. October 2019.

Banner Image Credits: Lung cancer cells, Anne Weston, Francis Crick InstituteCC BY-NC, Modified