How much is toxicity a function of the drug itself (eg small molecule, antibody, or -- stretching the meaning of drug here -- a particular gene edit) versus a joint function of the drug, formulation, delivery mechanism, etc? In other words, is toxicity something that can be answered at the target discovery stage, or will you always have to solve all the other steps and then check whether the final result is toxic? I realize the answer is probably fairly context dependent, but I'm curious whether there are broad trends among various drug, disease, or tissue types.
this is just one owls opinion but i suspect all the information you need to derive small molecule toxicity is available at the development stage
but for anything biologics-based, it's fuzzier, because the creation process behind the biologic starts coming into picture (e.g. antibody aggregation) + they often have more complex delivery mechanisms (e.g. LNP-based inflammation)
the real underlying problem is that biologists, chemists, bio-informaticists lack a causal model of any part of the body. so we rely on assays, 'expert' labels and shitty statistics for 30 years as a basis for a trillion dollar industry. this might take five years to play out (synthetic multi-level labels, consumer verbatims, raw input of in vivo experiments) but the entire field needs an enema (perhaps 'medical physics' will help). i say this as coming from the mobile space, where Jobs was nice enough to give us an enema, and society benefited
Yes, but the reason we don't have a model like that is because we don't have the ability to make one. Living things are insanely complicated at the level of biochemistry and it's also damn hard to monitor biochemistry in a living organism in real time - it's like trying to model an economy by simulating all the individual humans that make it up. It would indeed give you the right answers, but you can't actually do it!
Agree - data is hard to get, there is a Heisenebergian aspect that the act of collecting data disrupts the organism and therefore the data, and all that. Biology is hard, but so is Quantum Mechanics. My skepticism arises from working for years with folks who build Clinical test methods for stuff such as longitudinal testing of the efficacy of skin products (albeit could be any other longitudinal visual/sensorial study) - whose use of imaging and AI techniques is amateurish. It also comes from the gut reaction I've gotten on DL and GenAI methods from various kinds of scientists. Biologists are by far the most hostile. Even if CS people over-simplify things, they are incestuous in stealing and adapting ideas from other fields (and claiming it to be their own!) .. which leads to innovation velocity. By far I find Biology to be the strongest bunker mentality. I'm hoping blended companies like InSitro can create a culture that avoids trivializing science while at the same time bringing some hard core risk taking to what are the most important societal opportunities of our time.
How much is toxicity a function of the drug itself (eg small molecule, antibody, or -- stretching the meaning of drug here -- a particular gene edit) versus a joint function of the drug, formulation, delivery mechanism, etc? In other words, is toxicity something that can be answered at the target discovery stage, or will you always have to solve all the other steps and then check whether the final result is toxic? I realize the answer is probably fairly context dependent, but I'm curious whether there are broad trends among various drug, disease, or tissue types.
this is just one owls opinion but i suspect all the information you need to derive small molecule toxicity is available at the development stage
but for anything biologics-based, it's fuzzier, because the creation process behind the biologic starts coming into picture (e.g. antibody aggregation) + they often have more complex delivery mechanisms (e.g. LNP-based inflammation)
the real underlying problem is that biologists, chemists, bio-informaticists lack a causal model of any part of the body. so we rely on assays, 'expert' labels and shitty statistics for 30 years as a basis for a trillion dollar industry. this might take five years to play out (synthetic multi-level labels, consumer verbatims, raw input of in vivo experiments) but the entire field needs an enema (perhaps 'medical physics' will help). i say this as coming from the mobile space, where Jobs was nice enough to give us an enema, and society benefited
Yes, but the reason we don't have a model like that is because we don't have the ability to make one. Living things are insanely complicated at the level of biochemistry and it's also damn hard to monitor biochemistry in a living organism in real time - it's like trying to model an economy by simulating all the individual humans that make it up. It would indeed give you the right answers, but you can't actually do it!
Agree - data is hard to get, there is a Heisenebergian aspect that the act of collecting data disrupts the organism and therefore the data, and all that. Biology is hard, but so is Quantum Mechanics. My skepticism arises from working for years with folks who build Clinical test methods for stuff such as longitudinal testing of the efficacy of skin products (albeit could be any other longitudinal visual/sensorial study) - whose use of imaging and AI techniques is amateurish. It also comes from the gut reaction I've gotten on DL and GenAI methods from various kinds of scientists. Biologists are by far the most hostile. Even if CS people over-simplify things, they are incestuous in stealing and adapting ideas from other fields (and claiming it to be their own!) .. which leads to innovation velocity. By far I find Biology to be the strongest bunker mentality. I'm hoping blended companies like InSitro can create a culture that avoids trivializing science while at the same time bringing some hard core risk taking to what are the most important societal opportunities of our time.
For some reason I'm reminded of this XKCD:
https://xkcd.com/1605/