YSPH Biostatistics Seminar: “A Transfer Learning Approach to Build High-Dimensional Models Integrating Datasets with Disparate Features”

NOTE: BIS 526 students are required to attend in person. Others are invited to attend in person, but may also attend via Zoom.

Speaker- Nilanjan Chatterjee, Ph.D.

Title- “A Transfer Learning Approach to Build High-Dimensional Models Integrating Datasets with Disparate Features”

Abstract

Development of comprehensive prediction models are often of great interest in many disciplines of science, but datasets that have information on all desired features are typically small. In this article, we describe a transfer learning approach for building high-dimensional generalized linear models using data from a main study that has detailed information on all predictors, and from one or more external, possibly much larger, studies which have ascertained a more limited set of predictors. We propose using the external dataset(s) to build reduced model(s) and then “transfer” the information on underlying parameters for the analysis of the main study through a set of calibration equations that can account for study-specific effect of certain features, like design variables. We then use a generalized method of moment (GMM) with penalization approach for parameter estimation and develop highly scalable algorithms for fitting models with lasso and adaptive lasso penalties taking advantage of the popular GLM-net package. We further show use of adaptive-lasso penalty leads to oracle property of underlying parameter estimates and thus leads to convenient post-selection inference procedures. We conduct extensive simulation studies to investigate both predictive performance and post-selection inference properties of the proposed methods compared to traditional analysis of only the internal study. Finally, we illustrate a timely application of the proposed method for the development of risk prediction models for nine common diseases using the UK Biobank study, combining baseline information from all study participants (N=500K) and recently released high-throughout proteomic data (K=1500) on a subset (N=50K) of the participants.