Improving Reference-Free Isoform Reconstruction at Stockholm University
From August 19 to October 18, 2024, Fabian Jetzinger, LongTREC Doctoral Candidate at BioBam Bioinformatics, had the opportunity to carry out a research secondment at Stockholm University, supervised by Kristoffer Sahlin. This secondment offered valuable growth, both in scientific expertise and personal development.
My goal? Contributing to the development of isONform, a tool for reference-free isoform reconstruction from long-read sequencing data. The aims were to reduce redundancy in its output by collapsing isoforms with the same or highly similar open reading frames (ORFs), and to optimise a known computational bottleneck by re-implementing it in Cython.
Fabian reflects on his experience:
What I Learned ๐
Re-implementing a specific subroutine of isONform in Cython reduced the overall runtime of the program, with preliminary evaluation on synthetic data showing reductions of up to 55%. I also ran preliminary experiments using orfipy to predict ORFs from isONform’s results and collapse them into a less redundant set of isoforms.
Takeaways ๐ฑ
Early in the secondment, computational performance turned out to be a more significant issue than anticipated, so it was prioritised over the collapsing step. The initial ORF-based collapsing approach proved too aggressive, removing too large a portion of the transcriptome as indicated by LRGASP Challenge 3 evaluation. Time constraints prevented conclusive results on redundancy.
Impact ๐
Final evaluation of the performance improvements on real data and their deployment in a public release are planned for 2025.
Overall, this secondment delivered a faster isONform and a clearer view of what a collapsing step needs to avoid.
