 Do you see Fernando? We lost we lost participants but we found the model now we lost Fernando too. Fernando, whatever. So it's Ron here now it starts. Hey he is Ron. Hey Shailen, Alice. I am Shailen, not Absent. You are Shailen but did you see maybe we found the model? Yes, good. It didn't mean to be on plasmin but. Okay, good. So should we go to the last talk? Are there any more questions for Marco? Are there more questions? So otherwise we go for the last talk and then we see if there is another question or there is some discussion later on. Okay. So finally we have Ron. Oh no. Are you already? Can you share the screen? Yeah. So what's your face? Hey, hi. Nice to meet you. Okay, can everybody see my screen? Yes. Okay, great. So you have 15 minutes. Yeah, I'll try to keep it brief. So first of all, I had a weird crazy week and it was kind of a mess and I made two different time zone mistakes that messed up my previously two previously scheduled talk slots. So I really appreciate the organizers for being very flexible and generously and I'm really sorry for all the inconvenience I've caused and thanks for bearing with me and anyways here I am to do this talk finally. So what I'm trying to do is I'm trying to find interactions between genes in accessory genomes using patterns of gene gain and loss using presences and absences and absences of genes and this is very closely related to what James McInerney presented earlier about co-infinder and that's what history is doing. So for instance if we take and I apologize for coming into a plasmid workshop and immediately throwing up some linear genomes but anyways. So here's four examples of genomes that we've got these different colored genes of these solid colored genes are part of the accessory genome of these organisms and so you can see this pattern of blue and yellow these genes appear in when they appear they appear together well they don't appear they don't appear together and so that's one potential signal of interaction between these two genes. In this case it would be a what I would call a positive interaction that they either help each other or are required for each other's effect and then with this green and yellow or the green and blue genes we can see that the green gene is present when the other genes are absent and it's absent when the other genes are present and so that would be a signal potentially of some sort of negative interaction where it's either an alternative or a or a competitive gene and so there have been many methods that try to detect gene interactions using these presence absence patterns and so what I was trying to do was try to come up with a method that works for extremely more like on the order of thousands to tens of thousands of samples because this is an example tree that I'll show later based on the database I was using but a lot of a lot of times when you need to do co-evolution stuff you need to look at you need to look at the tree at least and try and often perform computations across the whole tree and when the trees get really big then this gets computationally intensive it's very slow and so the solution that I had for this is let's let's throw out most of the tree let's only look at the the thing that we think that I think will be the the strongest signal of genetic interaction in these accessory genomes which is uh whether or not there's differences between like the most closely related samples a pair of samples in the dataset so in general if you have closely related samples you don't expect them to differ very much and if they do differ that's interesting and then if they differ in the same way for different genes that's even more interesting and potentially tells us something about those genes so here's an example of three um three sets of pairs of individuals uh that here this a pair of closely related individuals here they these these individuals are identical and here in this pair of individuals one individual has these two genes the unoriginal individual doesn't and so what that would do is if if I see this pattern I chop that up to one I tell you that up uh for for one one point in the positive interaction store between the blue and yellow genes and if something looks like this with with these two close and real individuals where one of the individuals has the blue gene and doesn't have the yellow gene whereas the other individual doesn't have blue gene has the yellow gene then I chop that up as a point for a negative interaction between these two genes blue and yellow and then so I do that for every single uh pair of closely related individuals in the dataset and come up with a positive or negative interaction score for each pair of genes now so I played this method to a database collected by uh collaborators at emory university robert pettit and tim reed who compiled about 40 000 publicly available staph aureus genomes into this nice uh database called staphopia and so applied uh by a method to uh both um subsets and the whole dataset here and uh and this just this is that tree that I showed earlier this is just the tree of the samples so I applied my method to these samples and this is the the main result is like this is these these is the network of strongest interactions between genes in uh this database of staph aureus and so there's a couple of take home messages from these results here first um most of the strong interactions are either they're the color this red color here which uh is a signifies in the color scheme I was using virulence related genes so all these genes either code for some sort of toxin or some biofilm formation for host colonization or something else that involves that that has related to uh infection of the host with staph aureus as there's a couple of clusters of those virulence genes and then there's also the other prominent feature of this network is clusters of antibiotic resistance genes or antibiotic resistance genes in general in particular this cluster here of of which is a sec mech the concept that's the primary driver of methicillin resistance in staph aureus so all those are interacting with each other then there's also other antibiotic resistance genes here's some more beta-lactam resistance genes um and then we see that they're interacting with these cadmium resistance genes which uh reflects a known uh plasmid interaction um there and then there's a couple of other here this is a phosphomycin resistance gene and this is a another antibiotic resistance gene so the primary things that are happening in recent code genetic gene presence absence co-evolution in staph aureus are primarily driven by the host pathogen interaction whether it be virulence or resistance to antibiotics um and then it's all clearly obviously driven by the source only gene transfer and we see um certain genes like rap and rabae and uh my zoom thing is blocking one thing that was but I think it's pre these are our plasmid related genes so uh right so um the the line type here the dash the solid line indicates this is a positive interaction and the dashed line indicates that's a negative interaction and uh as you can see most of the lines here are solid which means there's a positive interaction this co-presence or co-absence um and the reason or and and so this is a consistent pattern across the entire days so this is here I've split up the data set into various different colore complexes which are a higher taxonomics uh a level for staph aureus than sequence type and so there's and um and then the y-axis is the fraction of the interactions uh that that are detected by my method that are positive and so you can see it across the board the majority of interactions are positive um which suggests that these that these accessory genes through horizontal gene transfer are um coming together into groups of genes that that work together and then that's the dynamic sort of driving this process and this result is um um consistent with previous results uh seen uh using different methods and different species like E. coli with uh uh coenfiner okay and so now this is a network of so the previous network was the strongest set of interactions in the whole data set this is the network of the interactions that are consistent of the strong and consistent across all the clinical complexes in the data set and so same similar story you can see there's fewer virulence-related genes here and more more resistance-related genes um and then um and then the mobilizing element genes are still here and so that's that that suggests that that the virulence evolution might be more environmentally specific whereas the resistance evolution has uh is consistent across the whole set of staph aureus genomes so you can also do this with uh you don't have to do this with with gene presence absence you can also do a binary trade uh any sort of binary trade so they have so anything you can code with a one or a zero uh you can apply this method to and so for instance so here I applied this to predicted antibiotic resistance phenotypes instead of specific gene presence absences in the same data set and on the x-axis uh here uh oh sorry in the lower triangle of this correlation matrix here is the score that you get the co-pollution score you get using my method and then the upper triangle here is just a straight up presence a correlation between the presence absence uh vectors of these antibiotic resistance phenotypes and so in so generally most of these results uh or um sort of the the patterns are mostly the same so in general we can see this there's this one big cluster of uh particular antibiotic resistances that tend to uh co have positive interactions and co-curve with each other they're driven by or they're centered on the beta-lactam resistances including the sec mech complex for that one the tetracycline has been of like size and nbls as well and then the less common and these are also the by far the most common antibiotic resistances present in the data set and the less common ones tend to form a sort of peripheral cluster which has high correlation but um they don't actually have much of a recent evolutionary signal suggesting that these uh antibiotic resistances um the relationships of due to our earlier splits or earlier events on the tree that my method which only looks at really close relatives won't pick up there and is specifically not supposed to pick up because um I don't want to confound uh the signal with phylogenetic difference and the other interesting thing about this is that phosphomycin resistance appears to be negatively associated with all these other ones and if we actually go back up to uh this network here you see this is a thought phosphomycin resistance gene and has a bunch of negative interactions with a bunch of other genes so I know it's going on with phosphomycin but it's doing something different for males in this dataset okay so I sort of blitz through that in an effort to save some time but in summary uh I've made a method to detect genetic interactions um in bacterial datasets on the order of thousands to tens of thousands of genomes and the the thing about this method is it doesn't require phylogenetic tree it just requires a distance matrix um and you don't have to traverse the tree all the time to uh to do anything with it and uh so we can see from the results that uh the reason of which is deforious in the accessory genome is primarily driven by host-passage interaction especially involving SCMEC and virulence the majority of interactions are positive which um again seems to concur with other results in different systems so maybe this is a general pattern of genome evolution in bacteria and so there's a preprint available online and the method that I'm currently writing in our package uh which is going to be called decoder detecting co-voluntary genes and relatives that should be available soon to enable this computation in general thanks for listening these are my advisors and collaborators and again thanks for pairing with me for uh my scheduling mishaps for this workshop okay thank you very much so I see Simone