Development of Statistical Software and Implementation of its Models for Operating Pharmaceutical and Genomics Data

概要

Statistics has great contribution to pharmaceutical and medical research. However, not so many pharmaceutical or medical researchers are familiar with the complicated statistical methodologies and programming. The powerful statistical calculation tools usually require programming skills, such as R. R is regarded as the standard statistical programming languages; however, it lacks a graphical user interface (GUI) that appeals to various users. Current GUIs built on top of R, such as EZR and R-Commander, aim to facilitate R coding and visualization, but most of the functionalities are still accessed through a command-line interface (CLI). To assist researchers of medicine and pharmacy in running the most routines in fundamental statistical analysis, an interactive GUI, MEPHAS, was developed to support various web-based systems that are accessible from laptops, workstations, or tablets, under Windows, macOS (and IOS), or Linux. In addition to basic statistical analysis, advanced statistics such as the extended Cox proportional hazard (CoxPH) model and dimensional analyses including partial least square regression (PLS-R) and sparse partial least square regression (SPLS-R) are also implemented in MEPHAS. An executable R package mephas was also implemented with the user-friendly GUI available under various computing environments.

Users can perform data analysis tasks such as managing data, analyzing data, and visualizing the results step by step without intensive statistical software training. MEPHAS covers adequate pharmaceutical statistics including statistical probability distributions and hypothesis testing with various data types. It also provides advanced statistical methods such as analysis with regression models and dimensional analyses. In addition, MEPHAS extends regression models to cover random or interactive effects and enable the prediction of the new dataset. MEPHAS made up for the lack of the AFT model in most statistical software and provided two types of time-to-event data analysis. It is also the first web-based application to produce dynamic results and 3D plots for PLS-R and SPLS-R methods.

PLS has been widely used in chemoinformatic data analysis. In the past decades, PLS also gained lots of applications in bioinformatic data. With the development of high-throughput technologies, gene expression data have become easier to obtain. Such data are enhancing our understanding of cancers and some other intractable diseases. Great concerns have been raised on using gene expression data to predict cancer patients’ survival time. However, as the most popular model to predict the survival times, the CoxPH model does not work on data with a great number of variables. To deal with this issue, researchers have developed various partial least squares (PLS) algorithm in CoxPH model to reduce the dimension of variables for survival prediction. As the extension of PLS and CoxPH model, the model, FiPLSCox, which combines forward interval PLS (FiPLS) and the CoxPH, was proposed and can be used to predict cancer patients’ survival outcomes.

PLS applied with CoxPH models for genomic data censored time were firstly discussed on both simulated gene expression data and real-world cancer data. The proposed FiPLSCox had competitive prediction performance compared with the previous PLSCox model and the prediction performance was assessed by the time-dependent AUCs and time-dependent prediction errors. The results indicated that this proposed FiPLSCox model had good performance for classifying new patients into clinically relevant high-risk or low-risk groups based on the gene expression and survival data from previous patients. Additionally, if the baseline hazard could be known, the final outcome from FiPLSCox can be used to predict patients’ survival probabilities, and furthermore to assist clinical physicians in making clinical decisions in the early diagnosis.