EVE General Discussion

Ok one more question, does it have to be only one answer per picture or is selecting several matches at once a possibility also ?

There is a reason why even researchers fight over this stuff and why using a computer actually is a better way as you can be far more objective.

The homogeneity is not very different from the examples given for nucleoplasm (especially if you look at the additional examples) and regular roundish nucleoli are clearly present.

Nuclear speckles are much more well defined points and none of the examples show such clear nucleoli holes.

do you know the majority of the correct ansewrs? cos if you dont have the majority then the incorect answer could end up correct if combined with the farmers and the button mashers

Summary of my replies:

#78

Difficult to tell for sure what is is when red/blue is toggled on but it looks like cytoplasm or aggresome to me (really ugly cells though :'( ). I guess the blue marker could hide a nuclear staining, but very hesitant reg. that.

#79

What you see in the upper left corner is a cell undergoing mitosis = dividing. The DNA (blue) has been duplicated and densed up, and the next step will be for the copies to be separated (by the red microtubules) into the two daughter cells.

#81

Nucleus and nucleoplasm are really similar, and it can be difficult for us as well to distinguish between them. Nucleoplasm = everything in the nucleus except the nucleoli, so I'd actually say it's a more specific term than nucleus. In some cell types, it's extremely hard to find examples of nucleoplasm, it just won't show in either the green nor the blue channel (for us, it's somewhat easier as we'll be looking at the same protein in three different cell types at the same time during the evaluation, and if one of them shows nucleoplasm, we will look more carefully for it in the others).

#82

As many as you see! I think (if I remember correctly) that approx 60% of all proteins show multiple locations. Most common is 2-3, but there are cases (although rarer) where one can see 3-4. I would be *extremely* hesistant to label more than 6 (which should be extremely uncommon).

#84

Humans are still so much better than computers (unfortunately...). But we're working on it ^^

#86

We have looked at all images (well, as I think images are uploaded continously, this is not perfectly true, but we will have looked at >99% of the images) and labeled them according to what we think we see. We are interested in what you think though, and whether you can spot patterns that we have missed out on.

Reg. farmers and button mashers, it will be easy to filter out those (according to our image analysis/programmer expert HPA_Dichroic), we do have a subset of very well classified images that we can look at to see who are just clicking randomly.

#87

From that image I agree it looks negative. Is that one of the known examples, since it's says it should be periphery (cell junctions?)? If so, I'd like to check up on it, it could be weeeak CJ, but I can't tell from this image.

Come join the dark side! (awww I'd like a thumbs up smiley!)

My accuracy rating was great on the test server, but on Tranquility I can't budge from 53%.

edit: it would be really nice if the example images also had an rgb filter.

So Discovery ... , quite a big challenge with really puzzling "tests"
I thought I could be good at that but unfortunatly some profond and obscur considerations prevent my accuracy to be better than 48% .
I was quite sure before submitting that I would get an incorrect match despise a really good odds
like this example http://imgur.com/JLX2vMj
It is still funny but don't know something missing ......

When below 50% you normally only get 'Known' samples, and the known samples normally only have a single attribute. Once over 50% you'll get a larger mix. Personally this threshold of known slides only should be much higher in order to make for better results on the unknown/unlabelled ones that use group consensus since it ensures that people working on the unknown slides actually know what they are marking.

While the random clickers can be filtered out of the results, they are having a significant negative impact in game due to being present in large enough numbers to skew everyone else's accuracy rating, meaning people are often being called 'wrong' for marking an actually existing feature simply because a large enough portion of people spammed cytoplasm on everything.

Agree that it's annoying that you get "wrong" despite you know you're right (I feel the need to make an more extensive tutorial video really explaining the difference between golgi/mitochondria/vesicles and mtoc/aggresome which ppl seem to have problems with).

For the examples I looked at when playing I don't think the farming seemed to be that bad. I could in most cases really understand why one would choose those locations (but of course there were some where I was like "coooome on... How can you click nucleoplasm when it's clearly fibrillar center and nucleus" )

Hopefully ppl who don't play it with intentions of being serious will quit soon.

I'm (everyone in the group) are super excited over the interest for PD, and the crazy number of ppl that have been playing and helped classified the images. We never imagined this kind of turnout.

Also, glad to hear you think they've improved somewhat already, crossing fingers it'll continue that way! (I've been home sick since Wedn so today is the first day I've been able to try it - am now SO annoyed at being stuck at 60% since ages... wtf? )