Zurich, Switzerland
Healthcare and Digitalisation.
- Refactored the original code so researchers can use the tool easier to deal with different tasks in RNAseq analysis. - Added new features to include triple-exon, isolated exon and short exon cases and found 200-500% more kmers. - Modularized and optimized the code to make it scalable to process more then 50GB gene data and to increase the running speed by 25%. - Rebuilt the interface and created four modes to allow the researchers to implement the whole analysis pipeline with Immunopepper, reducing 40% of the time on average. - Designed test cases for all modes and implemented unit testing to make further development more robust. - Released ImmunoPepper 1.0 and contributed to a manuscript introducing the software.
- Designed algorithms to verify ABB machine learning model’s robustness on given samples in less than 0.5 seconds. - Proposed new methods to train predictive models whose verification rate increased from 40% to 90%. - Improved the current published state of the art algorithm on LSTM model verification by 10% and only cost half of its time.