> Research & Grants
> Grants Program
> Research Grants
> Research Grants Awarded
Research Grants Awarded
Accurate Breast Cancer Prognosis
Background: Accurate breast cancer prognosis can spare a significant number of patients from receiving unnecessary adjuvant therapy. Some recent exploratory studies have demonstrated the potential value of gene signatures in assessing the risk of post-surgical disease recurrence. However, due to the limitation of computational algorithms used in the existing studies, the optimality and effectiveness of gene signatures are not well established. Moreover, these studies all attempt to use a genetic based prognostic system to replace the existing clinical rules while neglecting the valuable clinical information. Given the complexity of cancer prognosis, a more practical strategy is to utilize both clinical and genetic markers that may contain complementary information to each other. Objective: The objective of the proposed research is to develop an accurate breast cancer prognostic system by using a hybrid signature identified from genetic and clinical markers. Specific Aims: (1) to establish a breast cancer prognosis system through advanced machine learning algorithms based on the identified hybrid signature; (2) to evaluate the established prognostic system on publicly released breast cancer datasets; (3) to investigate the influence of the heterogeneity of breast cancer on prognostic performance; (4) to validate newly identified genes in independent tumor specimens. Study Design: The key challenge of the proposed research is feature selection capable of efficiently mining information from data with extremely large data dimensionality. We propose a new feature selection algorithm using optimization theories and numerical analysis techniques. Based on the identified hybrid signature, a breast cancer prognostic model will be developed by using advanced machine learning algorithms. Large-scale computational studies will be conducted to evaluate the performance of the constructed computational model against other existing prognosis approaches on publicly released breast cancer datasets. The influence of the heterogeneity of breast cancer on feature selection and prognostic performance will be investigated. The newly identified genes will be validated in independent tumor specimens using established, in-house techniques including micro-dissection and real-time, quantitative polymerase chain reaction assays. Potential Outcomes and Benefits of the Research: The success of the proposed research will be important in aiding physicians to make informed decisions regarding the necessity of adjuvant treatment, and may lead to the development of individually tailored treatments to maximize tumor regression and the efficacy of treatment. Consequently, this would ultimately contribute to a decrease in overall breast cancer mortality and a reduction in overall heath care cost. Furthermore, the target genes identified in the proposed research could ultimately serve as therapeutic targets, as there are currently no specific metastatic targets available.
Accurate breast cancer prognosis can spare a significant number of patients from receiving unnecessary adjuvant therapy (chemotherapy or hormonal therapy) and sufferings caused by the toxic side effects of the treatments. The current clinical systems perform poorly in assessing the risk of developing distant metastases in breast cancer patients. There is an urgent need for developing a more accurate prognosis criterion. After decades of research on breast cancer prognosis, many prognostic markers have been reported in the literature, including clinical markers and gene signatures. A critical question remains unanswered to date: what is the best we can perform in breast cancer prognosis given all clinical and genetic information by using advanced computational algorithms. In this project, we propose to conduct a large-scale computational study trying to answer the above question, and thereby to develop an accurate breast cancer prognostic system. The key challenge of the proposed research is feature selection capable of efficiently mining information from data with extremely large data dimensionality. We propose a new feature selection algorithm by using optimization theories and numerical analysis techniques. Based on the hybrid signature identified from both genetic and clinical data, a breast cancer prognostic model will be developed by using advanced machine learning algorithms. The performance of the constructed predictive model will be evaluated on publicly released breast cancer datasets by following a rigorous experimental protocol. Statistical data analyses will be performed to compare the predictive performances of the hybrid signature with existing prognosis approaches. An accurate breast cancer prognostic system would have a huge humanitarian and economic impact. We have conducted a preliminary study that clearly demonstrates the feasibility of the proposed research. Our results show that using a hybrid prognostic signature can improve the specificities of the existing gene signatures and the current clinical systems by more than 20% and 60%, respectively. In 2005, about 200,000 new cases of invasive breast cancer were diagnosed in the U.S., and most of them were recommended to receive adjuvant therapies. However, 70%-80% (approximately 150,000) of the patients who receive the adjuvant treatment would have survived without it. At the 70% specificity level for a prognosis system, which is possible as demonstrated in our preliminary studies, one can spare approximately 100,000 patients from the unnecessary adjuvant treatment and the unnecessary suffering. Moreover, these adjuvant therapies are expensive: 8 weeks of treatment can cost at least $20,000-$30,000. This means that our breast cancer prognostic system could potentially save approximately $3 billion in health care costs each year in the U.S. alone.