Data mining is the process by which valuable information is extracted from large datasets. As technology advances, the potential applications of data mining are being increasingly realized and explored; these applications include identifying potential targets, drug repurposing, and mutation-speciﬁc enrichment of established therapies.
One approach to drug repurposing in AML that has garnered attention is using machine learning to integrate big data for precision medicine. Data across multiple fields of study are being continually generated, highlighting the importance of determining what kind of data is the most predictive of therapeutically actionable events in cancer. In addition, it is still not clear what types of data integration will improve understanding of these disease networks.
Another approach involves inferring gene expression from epigenetic signatures and then matching these signatures to a drug-gene database in order to identify therapeutic targets. Network-based strategies for repurposing identiﬁed drugs that have antithetical effects compared with specific disease genes have also been employed.
A key limitation to data mining for mutation-speciﬁc therapies in AML is the lack of patient-matched genomic and clinical datasets. This can create challenges when investigating the effect of differences in therapeutic intervention on overall survival between patients with similar underlying mutations. A knowledge bank approach offers a potential solution.
Data mining is also being applied to extend druggable targets in AML by integrating different types of data and molecular profiles with actionable targets. Information derived from whole exome sequencing can be used to generate both tumor copy number information and mutation calls, the combination of which can inform inferences regarding complete or partial loss of function in a gene.
Data mining is additionally helpful in the search for novel therapeutic insights. Analyses of gene expression in AML have shown that overexpression or underexpression of correlated gene sets can distinguish different subsets of patients. Some gene sets can be directly linked to cytogenetic risk group and prognosis; others may be more reflective of the biological properties and differentiation state of blasts.
However, identifying mutation-speciﬁc therapies has proven difficult because the majority of mutations are likely “undruggable” and cannot be used as therapeutic targets. A promising alternative approach may be to identify therapies that prove harmful when a mutation is present, rather than targeting that mutation itself. The mutation and the second gene are then considered to be a synthetic lethal pair on the premise that a defect in either gene alone is compatible with viability, but the defects are lethal to the cell when they occur together.
“Any future bioinformatic-derived mutation-specific approaches must continue to be studied within the context of clinical trials but will need multicenter cooperation because of smaller and smaller molecular subgroups,” explained Dr Thomas. He noted that “one mutation, one drug approaches” are unlikely to improve outcomes in the long term.
“However, synthetic lethality is usually not taken into consideration by most hematology tumor boards,” he said. He added that combinations of mutations are usually not taken into account either.
“The good news is that even though it may seem overwhelming at first, good science and therapeutic creativity are going to be important in the next phase of precision medicine,” Dr Thomas concluded. “We are finally getting somewhere in a disease where very little has improved for many decades. The available therapeutic armamentarium and molecular tools to decide whether a patient will likely respond [to treatment] are the best they have ever been.”
- Benard B, Gentles AJ, Kohnke T, Majeti R, Thomas D. Data mining for mutation-specific targets in acute myeloid leukemia [published online February 7, 2019]. Leukemia. doi: 10.1038/s41375-019-0387-y
This article originally appeared on Hematology Advisor