What about AI doing our job in the future? https://www.nature.com/articles/d41586-019-01357-6?utm_source=Nature+Briefing&utm_campaign=4c1d57fdf3-briefing-dy-20190722&utm_medium=email&utm_term=0_c9dfd39373-4c1d57fdf3-44201949
Best Regards Nishant On Mon, 22 Jul 2019 at 11:30 PM, Sarah Bowman <[email protected]> wrote: > I'd like to point out that the MAchine Recognition of Crystallization > Outcomes (MARCO) makes a start to 'deep learning applied to crystallization > outcomes', at least in terms of being able to classify drop images > efficiently. > > > > There is obviously more work to be done to correlate these data with > crystallization cocktail components (which Janet and Tom point out the > difficulties with) and positive outcomes. It seems the first step really > needs to be consistent descriptions and vocabulary - I fully agree with > Janet here! > > > > Reference on MARCO for those interested: Bruno AE, Charbonneau P, Newman > J, Snell EH, So DR, Vanhoucke V, et al. (2018) Classification of > crystallization outcomes using deep convolutional neural networks. PLoS ONE > 13(6): e0198883. https://doi.org/10.1371/journal.pone.0198883 > > > > Cheers, > > Sarah > > > > *Sarah EJ Bowman, PhD* > > > > Associate Research Scientist, Hauptman-Woodward Medical Research Institute > > Director, High-Throughput Crystallization Screening Center > > Research Associate Professor, Department of Biochemistry, University at > Buffalo > > > > Research Webpage <https://hwi.buffalo.edu/scientist-directory/sbowman/> > > www.getacrystal.org > > > > [email protected] > 716-898-8623 > > > > > > *From: *CCP4 bulletin board <[email protected]> on behalf of Bernhard > Rupp <[email protected]> > *Organization: *k.k. Hofkristallamt > *Reply-To: *"[email protected]" <[email protected]> > *Date: *Monday, July 22, 2019 at 1:42 PM > *To: *"[email protected]" <[email protected]> > *Subject: *Re: challenges in structural biology > > > > What about 'deep learning' applied to crystallization outcomes? Can it > guide individual trials better than intuition? Can it find previously > unknown promising combinations on a larger scale? > > > > I think several people were well aware of this need for some sort of sound > machine learning already 15 years ago but we had no cloud based AI > > services then....maybe it is time to pick this up - particularly if face > recognition can classify the fine detail in faces maybe we finally could do > this with drop images as well... > > > > A summary of the state of affairs then is here: > > > http://www.ruppweb.org/cvs/br/rupp_2004_methods_predictive_models_crystallization.pdf > > > > LG BR > > > > > > Am 21.07.19 um 23:04 schrieb Artem Evdokimov: > > Dear Kay > > > > I disagree that 'magic bullet' is impossible. I think the definition is > wrong here - magic bullet to me is a rational set of methods that (when > executed with precision and care) enable crystallization to the maximum > possible benefit. This includes everything - constructs, crystallization > design, etc. Part of the magic bullet is also a precise knowledge when > crystallization is unlikely (i.e. an actual proven predictor that > consistently discriminates between "you're going to succeed if you work > hard" and "it's doomed to fail, don't bother" scenarios in crystallization. > > The above is not sexy. It does not present itself as a lovely subject on > which to have international cocktail parties with politicians delivering > fancy speeches. But that is what is needed, and no one is funding that to > the best of my knowledge. > > What needs to be done is a significant amount of testing, standardization, > and methods development from the perspective of holistic outcome (i.e. > crystals that work) - and none of the previously advertised 'magic bullets' > work the way I just described. > > Having written this, I think you're right - this is a bit of a distraction > from James' original point. However it's a valid opportunity for a lively > discussion on its own :) > > Artem > > - Cosmic Cats approve of this message > > On Sun, Jul 21, 2019 at 4:52 PM Kay Diederichs < > [email protected] <mailto:[email protected]> > <[email protected]%3e>> wrote: > > Dear Artem, > > black or white is not my way of thinking, which is why I don't > believe in Hannibal's approach when it comes to crystallization. > > None of the magic bullets that were advertised over the past decades > have proven generally applicable. I believe more in incremental > improvement which in this case includes a few biophysical characterization > methods, possibly improved microfluidics or other apparatus, and expanded > screens. And a lot of hard work, perseverance, intuition, frustration > > tolerance. Nothing that really needs huge funding - of course it > does need money, but just a share of what is anyway needed for the usual > lab work including expression, purification, functional characterization, > binding studies and the like. > > One area where a huge amount of money was burnt is crystallization in > space, on board of e.g. the spacelab and ISS. This is for me an example of > a mis-led approach to throw money at a difficult problem, with the > expectation of a solution. Science does not work like that, and money in > this case seems more to be the problem than the solution. > > This example may illustrate a certain failure of us scientists to > resist the temptation to promise unrealistic outcomes when confronted with > money provided for political reasons, which ultimately undermines our > credibility. But this takes us away from James' points. > > best, > > Kay > > On Sun, 21 Jul 2019 16:06:48 -0400, Artem Evdokimov < > [email protected] <mailto:[email protected]> > <[email protected]%3e>> wrote: > > >Dear Kay, > > > > > >Even the small, badly diffracting and 'messed up' crystals are still > > >crystals. There is literally a phase transition (pun very much > intended) > > >between growing *usable crystals* versus *having no crystals* (or > having > > >crystals that do not qualify as 'diffraction quality' even under the > most > > >favorable light). Points 2-9 fall into the 'I have crystals' bucket > and > > >everything else is in the 'I have no crystals' bucket. > > > > > >I am being deliberately black and white of course. > > > > > >As to whether huge funding would help to bridge the 'phase gap' - to > me > > >this is a purely theoretical question since to the best of my > knowledge > > >there never was a 'huge funding' for this particular problem :) And > if it > > >is true that the general belief in the art is that crystallization > is not > > >worth investing into because there's no hope in it then of course it > is a > > >self-fulfilling prophesy. > > > > > >There is an unresolved dichotomy buried in the sentiment above: it > seems > > >that we (the community of structural biologists) more or less > believe that > > >crystallization research is not fundamentally fruitful (hence the > > >no-funding situation). However, anyone who undertakes significant > efforts > > >to determine an actual structure using crystallography inevitably > *has to* > > >crystallize their target of interest - and therefore by definition > has hope > > >that their particular target will work out, against the overall > gloomy > > >outlook on the crystallization science as a whole. So we either are a > > >collective of self-induced schizophrenics, or the general sentiment > is > > >wrong and systematic crystallization research is meaningful and > > >fruitful - *just > > >very very hard*. > > > > > >In ~200 BC Hannibal reportedly said "I will find a way or make one". > I > > >think that if we approach problem #1 with this attitude (and an > equivalent > > >of a very large army's worth in funding) then it can be solved. > > > > > >Artem > > > > > >- Cosmic Cats approve of this message > > > > > > > > >On Sun, Jul 21, 2019 at 1:55 PM Kay Diederichs < > > >[email protected] < > mailto:[email protected]> <[email protected]%3e>> > wrote: > > > > > >> Hi Artem, > > >> > > >> you are certainly correct in that James' points 2-9 would be moot > if his > > >> point 1 were solved. But as long as this is not the case, we > resort to work > > >> with few and/or small and/or badly diffracting and/or > non-isomorphous > > >> crystals, which makes points 2-9 very relevant. > > >> > > >> Maybe the reason why crystallization research is not well funded > is that > > >> it is not expected to yield significant improvements. Personally, > I think > > >> that even huge funding would not result in methods that succeed in > > >> crystallizing all molecules. > > >> > > >> best, > > >> Kay > > >> > > >> On Sun, 21 Jul 2019 11:28:14 -0400, Artem Evdokimov < > > >> [email protected] <mailto:[email protected]> > <[email protected]%3e>> wrote: > > >> > > >> >Excellent question :) > > >> > > > >> >First of all, thank you for putting this out to the community! > > >> > > > >> >Secondly, I agree with several of us who've written that a single > > >> >conference is not enough to discuss all the possible topics. > > >> > > > >> >Thirdly, in my opinion all the other problems are secondary to > the main > > >> >(and only remaining!) problem in crystallography: getting > > >> >diffraction-quality protein crystals reproducibly and quickly > > >> > > > >> >The amount of funding for serious crystallization research seems > to be > > >> >close to non-existent. In general methodology funding is hard to > get, but > > >> >crystallization seems to me like the absolute underdog of the > method pool > > >> - > > >> >the true 'red headed stepchild' of the methods development > funders. > > >> > > > >> >At risk of repeating myself - the other problems (worthy, > significant, and > > >> >urgent as they are!) are subservient to the main issue at hand - > namely > > >> >that crystallization remains an unpredictable and artful > phenomenon while > > >> >literally all other aspects of structure determination process > (the gene > > >> to > > >> >structure pipeline, whatever you might call it)have made > astronomic leaps > > >> >forward. > > >> > > > >> >Artem > > >> >- Cosmic Cats approve of this message > > >> > > > >> > > > >> >On Mon, Jul 15, 2019 at 3:44 PM Holton, James M < > > >> >[email protected] < > mailto:[email protected]> > <[email protected]%3e>> wrote: > > >> > > > >> >> Hello folks, > > >> >> > > >> >> I have the distinct honor of chairing the next Gordon Research > > >> >> Conference on Diffraction Methods in Structural Biology (July > 26-31 > > >> >> 2020). This meeting will focus on the biggest challenges > currently > > >> >> faced by structural biologists, and I mean actual real-world > > >> >> challenges. As much as possible, these challenges will take > the form of > > >> >> friendly competitions with defined parameters, data, a scoring > system, > > >> >> and "winners", to be established along with other unpublished > results > > >> >> only at the meeting, as is tradition at GRCs. > > >> >> > > >> >> But what are the principle challenges in biological structure > > >> >> determination today? I of course have my own ideas, but I feel > like I'm > > >> >> forgetting something. Obvious choices are: > > >> >> 1) getting crystals to diffract better > > >> >> 2) building models into low-resolution maps (after failing at > #1) > > >> >> 3) telling if a ligand is really there or not > > >> >> 4) the phase problem (dealing with weak signal, twinning and > > >> >> pseudotranslation) > > >> >> 5) what does "resolution" really mean? > > >> >> 6) why are macromolecular R factors so much higher than > small-molecule > > >> >> ones? > > >> >> 7) what is the best way to process serial crystallography data? > > >> >> 8) how should one deal with non-isomorphism in multi-crystal > methods? > > >> >> 9) what is the "structure" of something that won't sit still? > > >> >> > > >> >> What am I missing? Is industry facing different problems than > > >> >> academics? Are there specific challenges facing electron-based > > >> >> techniques? If so, could the combined strength of all the > world's > > >> >> methods developers solve them? I'm interested in hearing the > voice of > > >> >> this community. On or off-list is fine. > > >> >> > > >> >> -James Holton > > >> >> MAD Scientist > > >> >> > > >> >> > > >> >> > ######################################################################## > > >> >> > > >> >> To unsubscribe from the CCP4BB list, click the following link: > > >> >> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > > >> >> > > >> > > > >> > >######################################################################## > > >> > > > >> >To unsubscribe from the CCP4BB list, click the following link: > > >> >https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > > >> > > > >> > > >> > ######################################################################## > > >> > > >> To unsubscribe from the CCP4BB list, click the following link: > > >> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > > >> > > > > > > >######################################################################## > > > > > >To unsubscribe from the CCP4BB list, click the following link: > > >https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > > > > > > > > > ######################################################################## > > > > To unsubscribe from the CCP4BB list, click the following link: > > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > > > > ######################################################################## > > > > To unsubscribe from the CCP4BB list, click the following link: > > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > > > > ------------------------------ > > To unsubscribe from the CCP4BB list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > ######################################################################## To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1
