วีดีโอ

Because this isn't an Apple bubble marketing ad, but an eyeopener what the right tool for the job can do.

You can use any windows laptop with mobile gpu and it will be a massacre

No problem, thanks for watching and let me know if you have any questions 🙂

No problem, and let me know if you have any questions! 🙂

Thanks and let me know if you have any questions! 🙂

No problem and let me know if you have any questions! 🙂

Ok, I'll try to make some more in the future!

Hi Harsha, sorry about the late reply! Unfortunately I don't really have a good answer to this question. Usually you have to kinda look around the other files they have posted, and try to find the one that has the relationship between sample labels and treatments. Or if you can't find it, try reading through the methods section of the paper to see if it's there. I purposely picked a nice and easy dataset to work with for this video series, but unfortunately sometimes it's not so easy to find all the information you need for the analysis. Good luck with your research!

Thanks for watching 🙂

Hi, sorry about the confusion! I put on the plot axis label that the unit here is in the hundreds, so the predator population was starting at 1 (hundred) and the prey population was starting at 10 (hundred). But I forgot to actually say this! So good point, yes if it was just one predator then it wouldn't make sense because they would't be able to reproduce.

@@MikeSaintAntoine if x=initial number of predators and y=initial number of prey do we always start by getting to a situation where x=1? it seems that this makes the calculations easier because x*y=y, am I thinking correctly?

@@ciferusbux Note that the values of x and y in this model are not necessarily integers. For example, x could be 327.61325.... The model isn't meant to provide an exact number of predator/prey at a certain time, it is meant to demonstrate the overall trends in the population change over time. Hopefully that helps.

Hi, off the top of my head I'm not sure what the error could be. If you've copied the exact same code I used and are still getting an error, that could be because we're using different versions of the same package, like maybe you have an older version, or maybe you have a newer version and they changed something so that my original code no longer works. Could you send me your code to take a look at? My email is mikest@udel.edu. Then maybe I'll be able to reproduce the error and fix it. Thanks!

Hi, yeah that's a good idea, I'll try to make a video on GSEA in the future!

Hey Zubair, good question and definitely not silly to ask! Yeah that part is a bit difficult to understand and it took me awhile to wrap my brain around it myself. I think a good first step is to understand exactly what we're trying to do when we randomly pick the next event. So basically we want to randomly pick the next event with a random draw, but with probabilities weighted according to the rates (also called "propensities"). And these rates change because they depend on the current level of X, so we need to calculate them again each time. So let's say that we're choosing the next event, and and we calculate that the rate for production is 2 (actually it will always be 2 in this example because it doesn't depend on X), and the rate for degradation is 1 (0.1 * X where X=10, for example). So knowing that the rate of production is 2 and the rate of degradation is 1, we want to randomly pick which event will happen next. So basically we convert these into probabilities by dividing the rate of each reaction by the sum of all the rates. So the probability that the next event will be production is 2 / (2+1) = 0.67. And the probability that the next event will be degradation is 1 / (2+1) = 0.33. So basically we want to choose the next event with a random choice, weighted according to these probabilities. The for-loops are just a way of doing that -- randomly picking a next event, weighted according to the appropriate probabilities. It's a little tricky to understand, but if you work your way through it, maybe on paper with the simple example, then you'll see that the first if-statement turns out to be true 67% of the time for the example I just gave, and the elif-statement turns out to be true 33% of the time. Note also that in this simulation with only 2 events we don't even really need the elif-statement. We could have just said "else" and left it at that. But I put the elif statement to show the general pattern, because it we had more than 2 events we'd need to put an elif-statements that follow that general pattern. Anyway, very good question. Yeah wrapping your head around the if-statement checks here can be a little tricky, but I think they key takeaway is that they're just picking the next event randomly, with probabilities that come from the relative rates of the reactions at this point in the simulation.

@@MikeSaintAntoine Ahhh I see. I think now I'm getting the gist of it. After playing around with different values for rand product on paper just like you said, I started to see how the value is weighted within the scope of probabilities of each reaction event. It's a bit tricky indeed but now I'm actually getting hang of it. Thank you so much for taking the time to break it down for me. As someone from medical biotech field it's not so very often for me to deal with such mathematical concepts. But you're making it digestible for us 😄. Please keep doing the good work!

@@zubairhasan9434 No problem, and let me know if you have any more questions! 🙂

Hi! Sorry but I don't think I really understand your question. Could you give me some more details on what you're trying to model? But generally a good way to see if this model applies to a dataset is to just plot the data and check it visually -- does it look like the population is increasing exponentially at first, but then leveling off at a carrying capacity?

Yeah there's a actually a hypothesis in economics that stock prices tend to follow a random walk: en.wikipedia.org/wiki/Random_walk_hypothesis so good observation! Thanks for watching 🙂

I'm not really in a good position to make an IDE video because I actually just use VSCode for writing Python, and don't really use any of the fancy extra features. I just use it like a text editor really. I know there are some pretty advanced IDEs for Python with built-in debuggers and stuff, but I don't really use them. But yeah in the future I can try to make some videos on dataframes / Pandas library. Thanks for watching! 🙂

Hey Arpit, yeah sure just email your code to me at mikest@udel.edu and I'll take a look!

Hi! Unfortunately I don't have a Windows or Linux computer so I won't be able to actually make a video on that. But my guess is that it's probably the same -- you should be able to install Python from python.org, and when you download and install it should give you the IDLE application automatically.

Yes, I'll try to make something like that in the future! I've actually already made a series like that in R, but I'd like to make one for Python too.

Hi, very sorry about the late response! Are you talking about "sample" in this block of code? for(sample in samples){ tmp <- unlist(strsplit(sample,"_")) classes <- append(classes, tmp[1]) } If so, "sample" is just the temporary variable we're using to iterate through the items in the "samples" vector. So basically it's getting assigned to each item in "samples" with each pass of the for loop. Does that make sense? I hope I understood your question correctly, but please let me know if you're still confused and I can try to explain further. Thanks for watching! 🙂

Yeah, good point!

Thanks for watching! 🙂

Yes, good point! If you're only doing neural networks, then the visited check isn't necessary. But part of the idea of the Value type is that it should be useful as a general gradient tracker for basic operations, which could include a cycle. Thanks for watching! 🙂

Yes, you're right! Good call on that one, that's an easier way of writing it than what I wrote.

Thanks for watching, and let me know if you have any questions! 🙂

Correct! Yeah this is called an "off-by-one" error and is a common bug that comes up a lot, both for beginners and even more more advanced programmers who make the absent-minded mistake.

@@MikeSaintAntoine Wow. Didn't know it's name. Thank you.

Good point! Yeah Python is more flexible with typing, and I think generally more convenient than R.

Hi! That's a good idea, I've been meaning to remake this video for both Python and Julia, but just haven't gotten around to it. But hopefully in the future once I have time. Thanks for watching! 🙂

Hi, sorry about the late response! I don't have much familiarity with LHS/PRCC sensitivity analysis, but if you email me your code at mikest@udel.edu I can take a look and try to help.

Oh nice, very cool! Yeah that's a good use of metaprogramming. I think it definitely makes the code cleaner and nicer, although with this series I was focused on teaching and readability rather than writing production-quality code, so I think for the goal of teaching it might still be easier to explicitly write out each function separately rather than use the metaprogramming approach. Thanks for watching! 🙂

Thanks for watching, and let me know if you have any questions! 🙂

Thanks for watching! 🙂

Hi, good question! Yes in this example it's looking at correlation for all samples across both conditions. So basically, a high positive correlation between two genes means that expression changed in the same way due to the drug (either both up or both down). A high negative correlation between two genes means that their expression changed in opposite directions due to the drug. Usually when I compute foldchange (or log2-foldchange), I do it like this: log2(mean(drug_expression) / mean(control_expression)), so that it gives only one number for each gene, since it's computed based on the mean expression in each condition. So with only one number per gene, you can't really compute the correlation between two genes. But another way to do it is to compute the log2-foldchange for each drug sample, relative to the control mean, so that will still give you a vector of numbers as a result and you could look at correlation between two genes. I'm not exactly sure if this makes sense though, since it seems like it's losing some information. Basically doing it this way is normalizing the drug-condition expression measurements by the control expression mean. So if we normalize away the effect of the drug, then we're losing that information when we eventually go to calculate the correlations between genes for the log2FC-transformed drug expression values. Does that make sense? If you need any help or want to discuss further, feel free to email me at mikest@udel.edu 🙂

@@MikeSaintAntoine@MikeSaintAntoine Hi, I am correlating gene expression values between two same genes (cross-talk genes) from different datasets, but each gene has a control and a case sample set. In this case, what value should I use for correlation analysis?

@@SNMS1114 Hi! I think in that case you should include both control and sample data, and compute the correlation for the genes across both of these conditions. That will give you an idea of how the case-condition is affecting their expression -- does it affect both in the same direction? or in opposite directions? or is there no correlation? If you need any help, please feel free to email me at mikest@udel.edu and I can take a look at your code. Thanks for watching! 🙂

I mean cross talk genes from the different case. But i want to perform pearson correlation analysis. So, each case has samples as control and case. When I compare the crooss talk genes from different two case.

Thanks for watching! 🙂

Good idea, I'll try to make a Python version when I have time!

@@MikeSaintAntoine 👍

Hi, sorry about the late response! Yeah that's a tricky problem with gene expression analysis. Often the datasets will give you some kind of ID number rather than the actual gene name. Is there another file somewhere in the dataset that tells you how the IDs correspond to gene names? If not, then you might need to figure out what type of IDs they are, and then you can use an R library called "org.Hs.eg.db" to convert them. For example, if the IDs look like this: ENSG00000139618, then those are called Ensembl IDs and you can convert them to gene names in R like this: library(org.Hs.eg.db) gene_names <- mapIds(org.Hs.eg.db, keys = gene_ids, column = "SYMBOL",keytype = "ENSEMBL") Anyway I hope that helps. Yeah these gene ID conversions are always a bit tricky and annoying.

Thanks for watching, and yeah I think explaining it as the amount of "surprise" a random variable has makes sense!

Thanks for watching and let me know if you have any questions! 🙂

What’s next??

Not sure! Ideally I'd like to add in some more functionality, to do some more complicated things like convolutions and transformers. Anything you'd like to see in particular?

Red- green.

Ah, I'm sorry to hear that! But maybe you can tell him to try to look on the bright side. Even though it limits what he can do, it won't disqualify him completely from joining like a lot of other medical conditions would. Also if he has any questions about it he can always send me an email at mikest@udel.edu.