To improve identification of canonical and non-canonical protein isoforms, we introduced ProteomeGenerator, a framework for reference-guided and de novo proteogenomic database generation from transcriptomic sequencing dataset. The proteomic databases output by ProteomeGenerator contain only proteins encoded by actively transcribed genes, and includes sample-specific protein isoforms resulting from non-canonical transcription and mRNA editing. We applied this workflow to the proteogenomic analysis of spliceosome-defective K052 SRSF2(P95H) cells, demonstrating high-confidence identification of proteins isoforms arising from intron inclusion and non-canonical splicing, as well as improved overall estimation of false-discovery rate from the focused database assembled by ProteomeGenerator.