The therapeutic use of synthetic message RNA (mRNA) has been validated in COVID-19 vaccines and shows enormous potential in developing infectious and oncological vaccines. However, double-stranded RNA (dsRNA) byproducts generated during the in vitro transcription (IVT) process can diminish the efficacy of mRNA-based therapeutics and provoke innate immune responses. Existing methods to eliminate dsRNA byproducts are often cumbersome and labor-intensive. In this study, we revealed that a loose mRNA secondary structure and more unpaired U bases in the sequence generally lead to the formation of more dsRNA byproducts during the IVT process. We further developed a predictive model for dsRNA byproducts formation based on sequence characteristics to guide the optimization of mRNA sequences, helping to minimize unwanted immune response and improve the protein expression of mRNA products. Collectively, our study provides novel clues and methodologies for developing effective mRNA therapeutics with minimized dsRNA byproducts and increased protein expression.