Abstract 4271: A new methodological approach to discovering biomarkers of mammographic breast density using pathway-guided lipid amalgamation

Abstract Introduction: High mammographic breast density (MBD) is a risk factor for breast cancer. Studies are evaluating the associations of various multi-omic biomarkers, in well-defined pathways with MBD, but the existing methodological approaches have drawbacks. For example, multivariate association analysis adjusting for covariates may identify too many biomarkers and lacks proper biological interpretation. On the other hand, simple amalgams (e.g., combination or sum) of species within pathways may result in few associated pathways, as the species within a pathway could have different directions of associations. Variable selection approaches such as the least absolute shrinkage and selection operator (Lasso) select individual species without considering prior knowledge of the pathway structure. We propose a new pathway-guided amalgamation method to form species clusters that could enhance biological interpretation. Methods: The plasma concentration of 982 lipid species in 14 super-pathways was measured in the blood samples of 705 premenopausal women drawn during their annual screening mammogram. We aimed to identify amalgams of lipid species that are associated with the volumetric percent density (VPD, in log scale). The proposed method utilizes the fused Lasso algorithm which penalizes the differences between effects for the lipid species within the same pathway and hence induces equality among similar effects. The lipid species with the same effects can be amalgamated to form clusters of species as new biomarkers. The number of clusters was chosen by cross-validation. Besides amalgamation, the method also facilitates variable selection by singling out influential individual species. Results: The proposed method identified 11 amalgams formed by 744 species, with 3 amalgams in Cholesteryl Esters (CE), and 1 amalgam each in Phosphatidylcholines (PC), Lysophosphatidylcholines (LPC), Phosphatidylinositols (PI), Dihydroceramides (DCER), Hexosylceramides (HCER), Lactosylceramides(LCER), Diacylglycerols (DAG), and Triacylglycerols (TAG). Amalgams in the same pathway can have opposite directions of association: 2 of the 3 amalgams in CE had negative associations with VPD while the third showed a positive association. Species with small effects in the same direction of associations may be combined: an amalgam comprising 518 species in the TAG pathway was identified with each presenting a small negative association with VPD but collectively, forming one amalgam with a large effect. Conclusions: The proposed amalgamation method provides a flexible and interpretable variable selection and clustering approach to discover lipidomic biomarkers for MBD. The amalgamation is especially preferred when multiple biomarkers species within one pathway share similar effects, or species within one pathway show opposite directions of associations. Citation Format: Chongliang Luo, Jingqin Luo, Kayla R. Getz, Myung Sik Jeon, Adetunji T. Toriola. A new methodological approach to discovering biomarkers of mammographic breast density using pathway-guided lipid amalgamation. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4271.