SpliceAI-blackbox

Using explainable AI to crack open the black box of SpliceAI

File structure

• motif_distillation/: Code to reproduce activation maximization
• motif_characterization/: Code to train surrogate multinomial model and localize motifs
• data-in/: Raw input data used by scripts
• data-out/: Intermediate processed data generated by scripts
• results/: Intermediate results generated by scripts
• SpliceAI-master/: SpliceAI model used by scripts

PWM Location

PWMs generated using activation maximization by gradient ascent are stored in results/A-ascend/

Citing

Sullivan et al, Manuscript in Prep

Contributions

• Patricia Sullivan: splicing expertise, data analysis, software development
• Dr Thom Quinn: ML/AI expertise, methods development, data analysis
• Dr Mark Pinese: splicing expertise, ML/AI expertise, genomic expertise, supervision
• A/Prof Mark Cowley: splicing expertise, genomic expertise, supervision