As high-throughput genomic sequencing generates increasing data volumes, researchers face challenges in identifying functional gene regions within DNA sequences. Traditional experimental methods prove time-consuming and costly for vast genomic datasets. The research addresses these challenges through signal processing frameworks, establishing automated mechanisms converting DNA sequences into numerical signals, enabling spectral analysis revealing structural features and periodic patterns characteristic of protein-coding regions.
The study introduces multi-scale spectral analysis combining the Discrete Fourier Transform for frequency analysis, Short-Time Fourier Transform and Gabor transforms for time-frequency localization, and wavelet transforms for multi-resolution decomposition. DNA sequences undergo numerical mapping, converting bases into discrete signals. Spectral methods extract trinucleotide periodicity characteristic of exons, while analysis windows capture features distinguishing functional regions. Support Vector Machine classifiers construct optimal hyperplanes employing kernel functions mapping nonlinear sequence patterns, enabling classification of gene segments versus non-coding regions.
Implementation validation incorporates testing on prokaryotic and eukaryotic genomic datasets, comparing optimized SVM classifiers against traditional approaches. Results demonstrated significant improvements through training set optimization, achieving 99.5% sensitivity with 5.89% specificity at optimized thresholds. Analysis across sensitivity levels showed SVM maintaining superior specificity, with 24.2% at 73.5% sensitivity and 18% at 82% sensitivity, effectively identifying functional fragments while minimizing false positives.
This research is led by Mr. Chongwei Shi, a biostatistics researcher at Georgetown University, with graduate-level training in biostatistics (University of Michigan) and a quantitative science background from UC Irvine, where he studied mathematics, data science, and economics. His technical work involves statistical computing, signal analysis, and machine learning using R, Python, and MATLAB. In addition to contributing to peer-reviewed venues and participating in academic peer review, he has developed two registered software tools that support biomedical data workflows, one of which has been licensed through a technology transfer agreement.
Mr.Shi has participated in research projects ranging from morphometric modeling in oral and maxillofacial contexts at the University of Michigan to functional genomics studies at UC Irvine, where he examined transcriptional and protein-level changes in yeast systems. Beyond genomics, his prior work spans disease survival modeling, stochastic processes, and quantitative economic analysis. This integration of signal processing and statistical learning demonstrates how computational approaches can accelerate genomic investigation without relying exclusively on wet-lab validation. The methodology supports automated gene region annotation at scale, offering practical utility for biomarker discovery and functional genomics. As sequencing datasets continue to expand, such analytical pipelines will play an increasingly important role in enabling efficient biological interpretation. Addressing key computational challenges in contemporary bioinformatics, these systematic approaches contribute to improved gene recognition accuracy and a deeper understanding of biological structure and function.