Democratizing statistical computing through AI

PROJECT SUMMARY Large data sets are routinely collected in biomedical research. Analyzing such datasets often requires coding, which can be challenging and time-consuming for researchers without computational expertise. Capitalizing on recent breakthroughs in artificial intelligence (AI), we developed a prototype of RTutor.ai, a website and a R package that enables users to "chat" with their data in dozens of natural languages. Under a reactive programming environment, we use OpenAI's ChatGPT to translate users' requests into R or Python code, which is then executed alongside user-uploaded data to produce results instantaneously. By integrating prompt engineering, code generation, execution, and reporting, RTutor enables people without coding experience to conduct preliminary analysis and visualization while boosting productivity for others by providing candidate code for further refinement. Despite limited features, the current version has been used by thousands of researchers world wild. This project aims to fully develop the prototype and rigorously test and document its capacities and limitations. Specific Aim 1: Develop and enhance RTutor. Based on our experience developing large bioinformatics apps such as iDEP and ShinyGO, we will (1) optimize prompts for effective and robust code generation, (2) upgrade to GPT-4 for better performance and improved code generation through conversations, (3) handle multiple datasets, (4) improve user interface, (5) automate exploratory data analysis, and (6) develop a customized version with built-in datasets and prompt structures for business customers. Specific aim 2: Test, evaluate and document. (1) Unit and system testing, usability testing by user groups with or without coding experiences, (2) systematic testing for limitations and robustness in biomedical data analysis, (3) test the applicability in various bioinformatics tasks such as RNA-Seq data, (4) document through text and video, (5) outreach to user communities. The completion of this project will lead to the development of an AI-powered virtual assistant for statistical computing. Like Amazon Alexa in our homes, its abilities might be limited but it can be made freely available to all scientists across the world, partially mitigating the shortage of experienced statisticians and bioinformaticians. RTutor has immense potential to accelerate scientific discovery and improve healthcare outcomes by enabling more researchers, especially those with limited resources, to gain insights from large datasets.