Ideas to discuss

[teresa-m]

Important:

• Which model should be our core model?
• What figure/table should be in the main text or supplement
• should we store the models/ datasets somewhere zenodo
• should we give a possible use case for Cherri. Where it can be helpful to solve a problem, close a gap?

For later:

• parralization of IntaRNA calles in pos/neg data
• is there a standard file format for RRIs we can use for Eval mode?
• a new method of hutter group to deep learning for tabular data could be used to build a model
• build in cross-validation for the model to also report an f1 score
• check memory consumption of feature selection + optimization and see if the parallel is an issue here
• homologs: Find homologes in human and mouse training data #32
• go away from genome-only sequences with context. e.g. give directly a mRNA
• update IntaRNA to genomic version IntaRNA

[teresa-m]

Possible applications

• postprocessing of sRNA, miRNA, gRNA,
• off-target predictions
• way to pre-filter RRIs for experiments

[teresa-m]

Tasks:

• get conda pacakge running
• Galaxy wrapper?
• update documentaion + paper on how to change IntaRNA paramteres
• Make the training data merge better usable
• Make a more generic input data possible (no need for Chira but it needs a header)
• Get functional testing/ GitHub actions
• Test the train and eval test calls (Maybe someone without prior knowledge)
• Affiliations of @martin and Egg
• Maybe:train models for single classes e.g. miRNA-mRNA, snoRNA-rRNA
• Maybe: Get LIGR-seq, SPLASH running. Or we take a CLIP-like method as an additional validation set.
• Maybe : Report feature importance