TransPrise

TransPrise predicts transcription start sites (TSS) in eukaryotic genomes using deep learning to determine precise TSS positions for genome annotation and transcriptional studies.

Key Features:

• Deep learning pipeline: Implements a two-part pipeline with a binary classifier to detect sequences likely to contain TSS and a regression model to determine the precise TSS location.
• Performance and accuracy: Reports a Matthews correlation coefficient (MCC) of 0.79 in Oryza sativa versus 0.31 for TSSPlant classification models, and regression mean absolute error (MAE) of 29.19 nucleotides for Oryza sativa and 47.986 nucleotides for Homo sapiens.
• Computational efficiency: Demonstrates GPU-accelerated processing, handling a 374 Mb genome in approximately 250 minutes.

Scientific Applications:

• Genome annotation: Provides precise TSS coordinates to improve annotation of eukaryotic genomes.
• Transcriptional regulation studies: Enables analysis of promoter architecture and transcription initiation by pinpointing TSS positions.
• Genetic resequencing and functional genomics: Supports analyses in resequencing projects and functional genomics that require accurate TSS localization.

Methodology:

Uses a deep learning two-part pipeline comprising a binary classifier followed by a regression model to localize TSS, with computation performed on a GPU.