Bioinformatic Evaluation of Transcriptomic Frame Selection Methods in Non-Model Species

Transcriptomic studies offer unique insights into differential gene expression, novel genes, and alternative splicing. In the absence of a reference genome, de novo assembly of the gene space proves challenging as organizing short reads is complicated by varying coverage and alternative splicing events. In non-model species where genetic resources for close relatives do not exist, the de novo assembly step is often followed by frame selection to identify the true start site and full open reading frame. This step is critical as it determines the translated protein and impacts the functional assignment of the assembled sequence. Similar to assembly, the existing bioinformatic tools for frame selection vary considerably when compared on the same read sets. Two current informatic methods for frame detection were compared in different non-model organisms with recent draft genome sequences including several members of Juglandaceae, Pseudotsuga menziesii, and Homalodisca vitripennis. The first method, Transdecoder, detects the longest open reading frame and optionally allows input of additional training data. The second, GeneMark S-T, employs iterative self training using an unsupervised machine learning approach. When compared across the Trinity assembled transcriptomes, GeneMark S-T provides consistent and accurate frame selection models within a reasonable run time. The study examined a variety of metrics, including: trimmed frame lengths, full-length annotations, accuracy of the translated frame, and alignments to the draft genomes. While comparisons of frame selection have been made on genes originating from model organisms, the results presented here provide the first comprehensive view of their impact on non-model transcriptomes.

Back to: Other Category

<< Previous Poster | Next Poster >>