4 ms·

My impression from reading The Computer Scientist’s Guide to Cell Biology is that the reason for woefully inadequate descriptions of cellular systems, rather th

My impression from reading The Computer Scientist’s Guide to Cell Biology is that the reason for woefully inadequate descriptions of cellular systems, rather than any fundamental deficiency on the part of biologists, is that the systems are so dang tiny and crowded, with every molecule running into every other molecule every second or so, at sizes literally below the widths of light waves, that we just don’t have the tools to physically observe what’s happening in real-time. Even the primitive equivalents of “shooting individual components” took so much clever innovation and hard work to figure out how to do that frankly it’s amazing that we know everything that we do.

nextos 4y ago
There are interesting attempts at modeling cells using Boolean networks (for example inferred using SAT/SMT), differential equations (preferably stochastic) and also with process algebra. Some are pretty successful. > [...] we don’t have the tools to physically observe what’s happening in real-time We do have single-cell technology, which is pretty close as you can get thousands of snapshots of different cells from the same lineage. One can then reorder them into some sort of pseudotime. There are also interesting real-time reporting systems for bacterial cells, but they are not high throughput. I think the field is still very young, and the current generation of biology professors is mostly allergic to formalism. Some departments are not, a notable exception are Caltech and Cold Spring Harbor (where the author of the essay is based). A related read is Luca Cardelli's followup: Can a systems biologist fix a Tamagotchi? http://lucacardelli.name/Papers/Can%20a%20Systems%20Biologist%20Fix%20a%20Tamagotchi.pdf http://lucacardelli.name/Papers/Can%20a%20Systems%20Biologis... And much deeper: Abstract Machines of Systems Biology http://lucacardelli.name/Papers/Abstract%20Machines%20of%20Systems%20Biology%20%28Draft%29.pdf http://lucacardelli.name/Papers/Abstract%20Machines%20of%20S...
- throwawayarnty 4y ago
  Single cell sequencing technology is analogous to taking many radios, and listing all the parts the radios contains. Then clustering radios by similarity in their part abundances. Useful, but a parts list is still far from figuring out how a radio works.
  - nextos 4y ago
    Not just that. For example one can perturb cells and try to understand their dynamics: https://cordis.europa.eu/project/id/810296 https://cordis.europa.eu/project/id/810296 Also, from purely observational studies it is possible to infer mechanistic models as I outlined above, e.g. using SAT.
Terr_ 4y ago
The author kinda touches on that with: > Another argument is that we know too little to analyze cells in the way engineers analyze their systems. But, the question is whether we would be able to understand what we need to learn if we do not use a formal description. The biochemists would measure rates and concentrations to understand how biochemical processes work. A discrepancy between the measured and calculated values would indicate a missing link and lead to the discovery of a new enzyme, and a better understanding of the subject of investigation. > Do we know what to measure to understand a signal transduction pathway? Are we even convinced that we need to measure something? As Sydney Brenner noted, it seems that biochemistry disappeared in the same year as communism (Brenner, 1995 ). I think that a formal description would make the need to measure a system's parameters obvious and would help to understand what these parameters are.
- joshuahedlund 4y ago
  Maybe it’s that I don’t grok what the author means by “formal description” but I don’t see how that really touches on it. As a layman the impression I got from the article is “biologists are doing the wrong things to learn about cells, I think they should be doing this other thing instead” whereas the impression I have from what I’ve read about those learning processes is that the learning is so hard that biologists are literally just doing whatever they can come up with to learn anything at all. They didn’t choose these processes because of what they wanted to learn, they literally invented those processes because it was what they could come up with (pretty ingeniously, too), there are no alternatives until those are invented and they would be ecstatic to use them as well. I probably have a disconnect from my lack of knowledge. An example would help; if it’s in the radio analogy I’ve lost it by the orders of magnitude difference in merely observing it at all.
SilasX 4y ago
>is that the systems are so dang tiny and crowded, with every molecule running into every other molecule every second or so, Per my two earlier comments [1], I would phrase it as an issue of biological systems optimizing soley for fitness, while (human-designed) computers are also heavily optimized for intelligibility, modularity, and ease of reasoning about. This makes it much easier to isolate and experiment with the subsystems and gain an understand, in contrast to biological systems, which will constantly bleed state across the entire system. [1] https://news.ycombinator.com/item?id=31710268 https://news.ycombinator.com/item?id=31710268 https://news.ycombinator.com/item?id=16818220 https://news.ycombinator.com/item?id=16818220

nextos 4y ago

There are interesting attempts at modeling cells using Boolean networks (for example inferred using SAT/SMT), differential equations (preferably stochastic) and also with process algebra. Some are pretty successful. > [...] we don’t have the tools to physically observe what’s happening in real-time We do have single-cell technology, which is pretty close as you can get thousands of snapshots of different cells from the same lineage. One can then reorder them into some sort of pseudotime. There are also interesting real-time reporting systems for bacterial cells, but they are not high throughput. I think the field is still very young, and the current generation of biology professors is mostly allergic to formalism. Some departments are not, a notable exception are Caltech and Cold Spring Harbor (where the author of the essay is based). A related read is Luca Cardelli's followup: Can a systems biologist fix a Tamagotchi? http://lucacardelli.name/Papers/Can%20a%20Systems%20Biologist%20Fix%20a%20Tamagotchi.pdf http://lucacardelli.name/Papers/Can%20a%20Systems%20Biologis... And much deeper: Abstract Machines of Systems Biology http://lucacardelli.name/Papers/Abstract%20Machines%20of%20Systems%20Biology%20%28Draft%29.pdf http://lucacardelli.name/Papers/Abstract%20Machines%20of%20S...

throwawayarnty 4y ago
Single cell sequencing technology is analogous to taking many radios, and listing all the parts the radios contains. Then clustering radios by similarity in their part abundances. Useful, but a parts list is still far from figuring out how a radio works.
- nextos 4y ago
  Not just that. For example one can perturb cells and try to understand their dynamics: https://cordis.europa.eu/project/id/810296 https://cordis.europa.eu/project/id/810296 Also, from purely observational studies it is possible to infer mechanistic models as I outlined above, e.g. using SAT.

Terr_ 4y ago

The author kinda touches on that with: > Another argument is that we know too little to analyze cells in the way engineers analyze their systems. But, the question is whether we would be able to understand what we need to learn if we do not use a formal description. The biochemists would measure rates and concentrations to understand how biochemical processes work. A discrepancy between the measured and calculated values would indicate a missing link and lead to the discovery of a new enzyme, and a better understanding of the subject of investigation. > Do we know what to measure to understand a signal transduction pathway? Are we even convinced that we need to measure something? As Sydney Brenner noted, it seems that biochemistry disappeared in the same year as communism (Brenner, 1995 ). I think that a formal description would make the need to measure a system's parameters obvious and would help to understand what these parameters are.

joshuahedlund 4y ago
Maybe it’s that I don’t grok what the author means by “formal description” but I don’t see how that really touches on it. As a layman the impression I got from the article is “biologists are doing the wrong things to learn about cells, I think they should be doing this other thing instead” whereas the impression I have from what I’ve read about those learning processes is that the learning is so hard that biologists are literally just doing whatever they can come up with to learn anything at all. They didn’t choose these processes because of what they wanted to learn, they literally invented those processes because it was what they could come up with (pretty ingeniously, too), there are no alternatives until those are invented and they would be ecstatic to use them as well. I probably have a disconnect from my lack of knowledge. An example would help; if it’s in the radio analogy I’ve lost it by the orders of magnitude difference in merely observing it at all.

SilasX 4y ago

>is that the systems are so dang tiny and crowded, with every molecule running into every other molecule every second or so, Per my two earlier comments [1], I would phrase it as an issue of biological systems optimizing soley for fitness, while (human-designed) computers are also heavily optimized for intelligibility, modularity, and ease of reasoning about. This makes it much easier to isolate and experiment with the subsystems and gain an understand, in contrast to biological systems, which will constantly bleed state across the entire system. [1] https://news.ycombinator.com/item?id=31710268 https://news.ycombinator.com/item?id=31710268 https://news.ycombinator.com/item?id=16818220 https://news.ycombinator.com/item?id=16818220