Preclinical Development – Chemical
2019 PharmSci 360
Genomics and related high-dimensional data are increasingly used in medical and pharmaceutical research for addressing various unmet needs related to early disease diagnosis, targeted treatment decisions, and optimal patient care. The recent advances in computing power, artificial intelligence (AI), and statistical/machine learning (ML) algorithms have provided a plethora of options for analyzing such data to develop prognostic and predictive biomarker signatures for diagnostic and precision medicine. An ideal biomarker signature that would be easier to implement in clinical practice would be in the form of a simple decision rule with a cut-point on each biomarker, similar to the way our lab results from the annual physical exams are still interpreted by physicians. Unfortunately in this world of AI and ML, most algorithms do not yield signatures in such simple forms; instead they are based on highly complex equations with invisible or uninterpretable structures that are often used like a black box. In this talk, we will review the application of our recently published algorithms (Chen et al, 2015, & Huang et al, 2017; Statistics in Medicine) for developing simple cut-point based biomarker signatures for a targeted lung cancer treatment (McKeegan et al, 2015, Lung Cancer) and for predicting Alzheimer’s disease progression (Devanarayan et al., 2019, Journal of Alzheimer’s disease). In addition to the simplicity of these signatures in the form of decision rules on just a few key biomarkers, we illustrate their superior performance over the more complex signatures in the literature.