/sci/ - Science & Math

>>16196813
ML in -omics is using fundamentally different datastructures to other tasks (hell, even GO term labels aren't particularly intuitive for prediction). Somewhere between financial transaction or social media temporal networks (where both use a fundamentally different definition of how time is represented) and NLP with far higher dimensionality.
You learn the underlying data structures/algos with compbio
You need to learn the statistics and underlying molecular biology to actually understand what's being answered (this goes into being able to decern what conclusions are appropriate depending on the experimental design)
these would be essential to develop the base genomic tools that wet lab scientists rely on, and fits in undergrad schedules. Moving on top of that can get you into epidemiology/med, working in hackathons for drug discovery or other techbro memes, a guaranteed paper name as the resident biostatistician, or learn how to graph and vectorize sequencing data for graphML, which is still a developing field for even natural network data.
Is GWAS bioinformatics fucking stupid? Yes, and you should not make that as a career goal, but there's a massive field in computational bio (and chem) that requires degree specialization and planning. You gotta find a niche, not "just take ML or Stats." You do that, and you become one of the 5000 others who post absolute crap on MDPI and shitjournals that wetlabs fucking hate. There is palpable disgust if you include the term multiomics in a presentation at physiology conferences because of the sheer volume of the crap that uninformed ML students in Tongji or Tianjin thinking they can model cancer.