Illuminating the dark side of the human transcriptome with long read transcript sequencing

The human transcriptome annotation is regarded as one of the most complete of any eukaryotic species. However, limitations in sequencing technologies have biased the annotation toward multi-exonic protein coding genes. Accurate high-throughput long read transcript sequencing can now provide additional evidence for rare transcripts and genes such as mono-exonic and non-coding genes that were previously either undetectable or impossible to differentiate from sequencing noise.

Long read transcript sequencing data has the power to identify novel genes within the highly annotated human genome. The use of parameter tuning and extensive output information of the TAMA software package allows for in depth exploration of eukaryotic transcriptomes. We have found long read data based evidence for thousands of unannotated genes within the human genome. More development in sequencing library preparation and data processing are required for differentiating sequencing noise from real genes in long read RNA sequencing data.

We developed the Transcriptome Annotation by Modular Algorithms (TAMA) software to leverage the power of long read transcript sequencing and address the issues with current data processing pipelines. TAMA achieved high sensitivity and precision for gene and transcript model predictions in both reference guided and unguided approaches in our benchmark tests using simulated Pacific Biosciences (PacBio) and Nanopore sequencing data and real PacBio datasets. By analyzing PacBio Sequel II Iso-Seq sequencing data of the Universal Human Reference RNA (UHRR) using TAMA and other commonly used tools, we found that the convention of using alignment identity to measure error correction performance does not reflect actual gain in accuracy of predicted transcript models. In addition, inter-read error correction can cause major changes to read mapping, resulting in potentially over 6 K erroneous gene model predictions in the Iso-Seq based human genome annotation. Using TAMA's genome assembly based error correction and gene feature evidence, we predicted 2566 putative novel non-coding genes and 1557 putative novel protein coding gene models. Conclusions: Long read transcript sequencing data has the power to identify novel genes within the highly annotated human genome. The use of parameter tuning and extensive output information of the TAMA software package allows for in depth exploration of eukaryotic transcriptomes. We have found long read data based evidence for thousands of unannotated genes within the human genome. More development in sequencing library preparation and data processing are required for differentiating sequencing noise from real genes in long read RNA sequencing data.