As genomics research accelerates, scientists face challenges in identifying functional proteins and understanding biological regulation. Traditional prediction methods relying on manual feature engineering struggle with complex sequence processing, resulting in limited accuracy. The research addresses these challenges through deep learning frameworks, establishing automated feature extraction mechanisms that capture both local sequence patterns and long-range dependencies in biological data. This research area has become strategically important within the United States as biostatistical modeling and computational genomics now support precision medicine, drug discovery, and population-scale disease surveillance. Federal initiatives such as the NIH All of Us Research Program, the Cancer Moonshot, and the CDC Genomics and Precision Health Program rely heavily on scalable statistical and computational models to interpret biomedical data.
The study introduces hybrid neural network architectures combining convolutional layers for local feature extraction with bidirectional Long Short-Term Memory networks for contextual learning. DNA sequences undergo k-mer encoding, generating high-dimensional vectors capturing nucleotide relationships. Convolutional operations extract conserved sequence patterns and binding site motifs, while BiLSTM layers process bidirectional sequence information, capturing long-range interactions, with fully connected layers performing binary classification.
Implementation validation incorporates testing on multiple benchmark datasets, comparing proposed architectures against traditional approaches. Results demonstrated 93% accuracy, significantly outperforming Support Vector Machines at 85% and Random Forests at 87%, achieving 91% precision and 92% recall with F1-scores of 0.915. Parameter optimization across varying k-mer lengths, LSTM configurations, and convolutional filter numbers established optimal architectural designs for biological sequence analysis. Industry analysts identify biomedical data analytics as a core growth engine for the U.S. biotechnology and life sciences sectors, which collectively contribute over two trillion dollars annually to the national economy. Methods that improve genomic feature extraction, disease gene identification, and protein function prediction directly support U.S. pharmaceutical R&D pipelines and molecular diagnostics development. These applications demonstrate that computational biostatistics has become an enabling technology rather than a purely academic discipline, with stakeholders spanning industry, research laboratories, and federal health agencies.
Contributing to this research is Chongwei Shi, currently pursuing a Ph.D. in Biostatistics at Georgetown University, holding a Master of Science in Biostatistics from the University of Michigan, and dual bachelor’s degrees in Mathematics with Data Science concentration and Quantitative Economics from UC Irvine, where he earned Dean’s List honors. Technical expertise spans R, Python, and MATLAB for statistical computing and data analysis. Academic achievements include peer-reviewed contributions for Applied Computational Intelligence and Soft Computing, Genetic Epidemiology, Engineering Optimization, and Journal of Statistical Computation and Simulation, demonstrating recognized expertise in computational methods. Two registered software copyrights, including Biological Statistics Data Analysis Optimization Management System and Genotype-Phenotype Association Platform, led to a technology transfer contract with Beta University. Shi’s published work has accumulated more than 57 citations with an h-index of 5, indicating that researchers in computational genomics and biomedical data science actively build upon his findings. In addition to scholarly publications, Shi has developed two registered biostatistical software platforms that support genomic data analysis and phenotype association studies. Such tools address common bottlenecks in contemporary biomedical research, including dataset integrity, statistical reproducibility, and analytical scalability. Shi has also served as a peer reviewer for Engineering Optimization, a Q1-ranked journal (CiteScore Best Quartile) with a 6% acceptance rate, reflecting independent recognition of his technical judgment.
Professional research at the University of Michigan as a Research Assistant for Oral and Maxillofacial Surgery applied Procrustes analysis to rat mandible morphometrics, utilizing MATLAB for landmark extraction from 3D scans while employing R and Python for statistical modeling. Experience at Zhang Lab of Molecular & Genome Evolution analyzed gene functions across yeast species, investigating protein and mRNA changes from gene knockouts through differential expression analysis. Additional contributions span survival analysis for disease prediction, stochastic process applications, and econometric analysis.
The integration of deep learning research with biostatistical applications demonstrates how computational frameworks translate into biological discovery. By establishing automated protein function prediction methodologies while deploying statistical shape analysis supporting precision medicine applications, this work bridges theoretical innovation with practical biomedical value, addressing computational challenges facing genomics and morphometric analysis through systematic approaches delivering improvements in biological understanding and clinical applications. As clinical genomics and precision medicine continue to expand, computational models of biological sequences are expected to become increasingly central to U.S. biomedical innovation.