Ideas for further development

[jdwinkler]

• RNA-level predictions: free energy changes associated with nucleotide changes within a window (X bp before, X bp after) a nucleotide change; screen for RNAse targets within sequences (not sure if that is possible)
• Protein-level predictions: residue conservation in proteome for every protein coding gene; structural information from AlphaFold2, Phyre2, or the like; protein-protein contact residue annotations; predicted DNA binding sites for DNA binding proteins or (anti)terminators; protein-ligand binding for mutated versus WT strains
• Ligand predictions: record known/general protein-compound interactions to help generalize effect of mutations, especially pumps etc
• Regulatory predictions; avoid treating the regulatory network as a singular entity until a sufficiently accurate model is developed, but try to understand the impact of global regulator mutation using each phenotype's known set of genotypes; changes in TF binding preferences upon mutation

Higher level:

• Hypothesis generation: given the available data, can we automatically identify a hypothesis that explains how the phenotype of interest arises?
• Same species library design: can we design targeted library experiments that appear likely to generate the resistance phenotypes in a WT E. coli strain?
• Cross species library design: can we design targeted library experiments to engineer the phenotypes of interest into another species based on our inferred molecular mechanisms and homologs in strain 2

[hypnopump]

Hi there! Very cool & useful repo!

I'm planning to build a model for genotype2phenotype mutation effects (mainly incorporating NLP and geometric deep learning models), and i have thought of similar points as the ones you outline! It could be interesting if we could try one of these models ( https://github.com/hypnopump/ClynMut/ ) for microbial resistance! I'm also taking part in the development of an open source AlphaFold2 replication, so we could get the structural information from there once it's ready (estimate 1-2 months for an open AF2).

I'd love to talk more on the possibilities of such a project!

[jdwinkler]

Great! Glad you think it's useful. I'll be happy to elaborate over email, so just get in touch with me at james.winkler@gmail.com!