101 video feedback

Feedback & Feature Requests
1

Hi, I went through the 101 video by Google DeepMind. As this video is from Google DeepMind, I was expecting a little detailed explanation of how to use AlphaGenome meaningfully. But unfortunately it was a such a basic video.

For example -
the person is just explaining like: this is the input sequence, this is the variant, this is the plot you get etc etc. I mean this explains nothing. This much anyone can understand by just looking at the code in the ‘quick_start’. Nowhere you are explaining what a certain block of code is doing, what are the lines in the plots, what does plots show, how to interpret the plots

As a team behind AlphaGenome, I think you should have made the video more detailed and explanatory.

Thankyou

2

Hi Anjali,

Thanks for sharing your feedback. The video was meant to be very basic and introductory.

Could you please further expand on what kind of content you would like to see in the video?

Best

Ziga

3

Thankyou for responding sir,

I would like to know more about how to interpret the plots of different modalities?

I was using SNPs, to understand the impact of SNPs. In ‘quick_start.ipynb’ the plots you have provided, there are clear differences between ref and alt signals but in my case for all of my SNPs, I got no difference in the signals. Ref and alt are completely superimposed on each other.

So I would like to understand the plot outputs of different modalities and their implications.

Looking forward to it

4

Got it. We can try to record another video focused on workflow recommendations and interpretation. Let us know if anything else comes to mind.

For your use case, I would recommend is that you first run score_variant(s) with quantile normalisation enabled. If you are seeing that all the quantiles are below 0.99, then it’s not even worth making the plots because they will all have the same value. You’d only want to look at the plots if you see something that is ‘out of distribution’, i.e. >0.99 quantile.

One other thing I would recommend is using Gemini or other LLMs to help you with the interpretation of the results. You can use the AlphaGenome documentation and/or different parts of the paper as additional context.