Robust analysis of high-dimensional omics data using computer simulation and graphical visualization

Abstract

In recent years, scientific research has faced significant challenges, including the rise of "fake news" on social media, which has complicated the public's perception of scientific truth. This is juxtaposed with the scientific principle of falsification, where hypotheses are not proven but rather refuted. Additionally, the reproducibility crisis—where many scientific findings cannot be consistently replicated—has become a pressing issue, highlighted in journals like *The American Statistician*. The discourse around p-values, with some suggesting their abandonment, underscores the call for more robust statistical analyses to ensure trustworthy scientific conclusions. This work explores methods to validate and enhance the robustness of statistical techniques for their effective application in real-world data scenarios. In clinical research, the objective of translating findings from the laboratory to patient care is crucial. This is structured by the 4T model: starting from basic research (T1), moving to evidence-based guidelines (T2), then transforming into clinical practices (T3), and finally improving community health outcomes (T4). Throughout these stages, biometry and statistical bioinformatics play a vital by developing robust methods that can validate clinical research findings. For example, the median offers a robust measure of central tendency less affected by outliers compared to the mean, thus providing more reliable results. The era of "big data" has led to unprecedented data volumes, posing new challenges in data management and analysis, and giving rise to the field of data science. In biometry, big data challenges were first encountered with genetic microarray technology, presenting the "high-dimensional data problem" where the number of parameters exceeds the sample size. This is illustrated in genome-wide association studies (GWAS), where genetic data consisting of hundreds of thousands of variants are analyzed to predict and classify diseases such as rheumatoid arthritis, demonstrating the necessity for innovative data processing and analytical methods to manage such complex datasets effectively.