According to the study, the predictive model developed with the use of machine learning can identify African-American patients with breast cancer with a high mortality risk.
"By using gene expression data, we developed a model of machine learning to accurately distribute African-American patients with breast cancer with high and low risk of death, - says Shristi Bhattarai from the State University of Georgia. - Since African American women tend to be the worst results of breast cancer, this study will help us identify the racial differences in this cohort and to develop personalized therapy specifically for African-American women. "
We African American women with breast cancer mortality rates adjusted for age, 40 percent higher than in Caucasian women. The etiology of this disparity results due to a number of factors.
Using data Atlas cancer proteome, Bhattarai and his colleagues have analyzed the protein levels of expression of 224 proteins in 754 patients with breast cancer. Of these patients, 620 were of European descent, and 134 were African American. The authors developed an algorithm to identify protein combinations that were associated with survival in breast cancer.
deep learning algorithm identified the combination of four proteins for optimal forecast. This combination will help distribute African-American patients with high-risk medical history with 86 percent accuracy.
"It is interesting that these proteins alone had no significant predictive value, - said co-author Sergei Klimov lab Rita Aného at the Department of Biology at the Georgia State University. - However, their combined effect in machine learning models could identify African American cohort, in which the mortality risk is five times higher. "
After controlling for clinicopathological parameters, including patient age and stage of cancer, the model could identify African-American women, whose risk of mortality is 11 times higher.
The researchers were unable to identify the European-American patients with breast cancer in a population with low and high risk with this model, suggesting that this it is forecasting only for African-American patients with breast cancer.
"We plan to start clinical trials in which we can identify very specific patterns of different demographic groups and to find patients at high risk to develop additional therapies - says Aného. - We are pleased that our model has the potential for relevant clinical trials. In the future, we will need to confirm this model in different groups of African-American patients with breast cancer. "