High-throughput technologies, such as magnetic resonance imaging (MRI) and DNA/RNA sequencing (DNA-seq/RNA-seq), have been increasingly used in large-scale association studies. With these technologies, important biomedical research findings have been generated. The reproducibility of these findings, especially from structural MRI (sMRI) and functional MRI (fMRI) association studies, has recently been questioned. There is an urgent demand for a reliable overall reproducibility assessment for large-scale high-throughput association studies. It is also desirable to understand the relationship between study reproducibility and sample size in an experimental design. In this study, we developed a novel approach: the mixture model reproducibility index (MMRI) for assessing study reproducibility of large-scale association studies. With MMRI, we performed study reproducibility analysis for several recent large sMRI/fMRI data sets. The advantages of our approach were clearly demonstrated, and the sample size requirements for different phenotypes were also clearly demonstrated, especially when compared to the Dice similarity coefficient (DSC). We applied MMRI to compare two MRI or RNA sequencing data sets. The reproducibility assessment results were consistent with our expectations. In summary, MMRI is a novel and useful approach for assessing study reproducibility, calculating sample sizes and evaluating the similarity between two closely related studies.
bioRxiv Subject Collection: Neuroscience