Memory-Efficient Isoform Clustering at CSIC
From December 3, 2024 to February 24, 2025, Mahmud Sami Aydin, LongTREC Doctoral Candidate at Stockholm University, had the opportunity to carry out a research secondment at CSIC, supervised by Ana Conesa. This secondment offered valuable growth, both in scientific expertise and personal development.
My goal? Exploring mapping-friendly compression methods to create a memory-efficient isoform clustering algorithm.
Sami reflects on his experience:
What I Learned ๐
We compared two mapping-friendly compression approaches for mapping RNA reads to a transcriptome: MSR (Mapping-friendly Sequence Reduction) and our own Syncmer Digest tool. Syncmer Digest achieves higher compression, but MSR is the optimal accurate solution for halving the dataset and gives better accuracy.
Takeaways ๐ฑ
We tried different hash functions to improve accuracy, but at higher compression rates Syncmer Digest could not reach MSR’s accuracy.
Impact ๐
We then developed a memory restriction mode for isONclust3, an isoform clustering algorithm, using external memory sorting and batching. It uses less memory than the default mode without losing the algorithm’s speed advantage, making isoform clustering more accessible on computers with less memory.
Overall, this secondment turned an exploration of compression methods into a practical memory-saving feature for isONclust3.
