TY - GEN
T1 - Predictive Signatures for Lung Adenocarcinoma Prognostic Trajectory by Multiomics Data Integration and Ensemble Learning
AU - Lee, Hayan
AU - Feng, Gilbert
AU - Esplin, Ed
AU - Snyder, Michael
N1 - Publisher Copyright:
© 2021, Springer Nature Switzerland AG.
PY - 2021
Y1 - 2021
N2 - Lung cancer is the most prevalent cancer worldwide. About 80% to 85% of lung cancers are non-small cell lung cancer (NSCLC). One of the major types of NSCLC is lung adenocarcinoma (LUAD), which solely accounts for approximately 40% of all cases. Although there has been a dramatic therapeutic improvement, the prognostic trajectory has relied on primarily clinical features such as tumor-nodal-metastasis (TNM) stage, age upon diagnosis, and smoking history for decades. It does not reflect molecular alterations on its pathway or heterogeneity of tumorigenesis. Here we propose an integrative multi-omics random forest model to predict survival time for LUAD patients. We identified multi-omics signatures with higher importance to better predict survival time than clinical annotations that physicians traditionally use. We confirmed that the integrative prediction model outperforms any single-omic-based model. We discovered that a methylation-based model performed best among any single-omic-based model for LUAD since it provides the most abundant signature candidates. Although methylation assay is costly in general, paradoxically, methylation offers the most economical pool as prognosis markers due to more abundant assay points.
AB - Lung cancer is the most prevalent cancer worldwide. About 80% to 85% of lung cancers are non-small cell lung cancer (NSCLC). One of the major types of NSCLC is lung adenocarcinoma (LUAD), which solely accounts for approximately 40% of all cases. Although there has been a dramatic therapeutic improvement, the prognostic trajectory has relied on primarily clinical features such as tumor-nodal-metastasis (TNM) stage, age upon diagnosis, and smoking history for decades. It does not reflect molecular alterations on its pathway or heterogeneity of tumorigenesis. Here we propose an integrative multi-omics random forest model to predict survival time for LUAD patients. We identified multi-omics signatures with higher importance to better predict survival time than clinical annotations that physicians traditionally use. We confirmed that the integrative prediction model outperforms any single-omic-based model. We discovered that a methylation-based model performed best among any single-omic-based model for LUAD since it provides the most abundant signature candidates. Although methylation assay is costly in general, paradoxically, methylation offers the most economical pool as prognosis markers due to more abundant assay points.
KW - Integrative multi-omics model
KW - Machine learning
KW - Survival time prediction
UR - http://www.scopus.com/inward/record.url?scp=85121915950&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-91241-3_2
DO - 10.1007/978-3-030-91241-3_2
M3 - Conference contribution
AN - SCOPUS:85121915950
SN - 9783030912406
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 9
EP - 23
BT - Mathematical and Computational Oncology - Third International Symposium, ISMCO 2021, Proceedings
A2 - Bebis, George
A2 - Gaasterland, Terry
A2 - Kato, Mamoru
A2 - Kohandel, Mohammad
A2 - Wilkie, Kathleen
PB - Springer Science and Business Media Deutschland GmbH
T2 - 3rd International Symposium on Mathematical and Computational Oncology, ISMCO 2021
Y2 - 11 October 2021 through 13 October 2021
ER -
