These forums have been archived and are now read-only.

The new forums are live and can be found at https://forums.eveonline.com/

Issues, Workarounds & Localization

 
  • Topic is locked indefinitely.
 

Project Discovery Matching Issues.

First post
Author
Previous Topic Next Topic
Rengaros
Shattered Void
#1 - 2016-03-30 02:35:21 UTC
I think it's great that something like this has been implemented that can have an actual impact on real science. I don't know to what extent the protein mapping matters or is trusted by the experiment but there are extreme inaccuracies in the "mini-game". If the results we get from the mapping does not matter, then that's great but if the Right/Wrong results do matter in some way to what we are doing then there are some real issues.

More often then not, the cells we observe are completely different than the actual "correct" answer which we get. For example, cells with nuclear bodies which means they are completely and densely filled with the green hue are categorized as nuclear speckles which have only partial cover of the nucleus. Also, cytoskeletons are commonly categorized as cytoplasm and Cell to Cell variation does not function properly at all, cells often present clear variations but are never categorized by the system as having them.

These are the most common I have come across so far. Again, if the percentage we are shown as right or wrong does not matter to the results that are used in the research then these are not really issues at all. On the other hand, if it does matter, it can provide with very inaccurate results.

Is there anything to be done to fix this, do these results on our end matter at all?
HPA Dichroic
H P A
C C P Alliance
#2 - 2016-03-30 07:18:16 UTC
Rengaros wrote:
I think it's great that something like this has been implemented that can have an actual impact on real science. I don't know to what extent the protein mapping matters or is trusted by the experiment but there are extreme inaccuracies in the "mini-game". If the results we get from the mapping does not matter, then that's great but if the Right/Wrong results do matter in some way to what we are doing then there are some real issues.

More often then not, the cells we observe are completely different than the actual "correct" answer which we get. For example, cells with nuclear bodies which means they are completely and densely filled with the green hue are categorized as nuclear speckles which have only partial cover of the nucleus. Also, cytoskeletons are commonly categorized as cytoplasm and Cell to Cell variation does not function properly at all, cells often present clear variations but are never categorized by the system as having them.

These are the most common I have come across so far. Again, if the percentage we are shown as right or wrong does not matter to the results that are used in the research then these are not really issues at all. On the other hand, if it does matter, it can provide with very inaccurate results.

Is there anything to be done to fix this, do these results on our end matter at all?



So the short answer is yes, the results on your end matter. To see some hard numbers checkout the HPA blog I wrote on day 1 results. Although there are common confusions, in general people do pretty well.

The more in-depth answer is that we're still looking into it. I'm going through the data now and trying to figure out how big these types of issues are and if there is some easy filtering that can be done (like requiring a higher threshold of significance for cytoplasm for example).

We have also changed the incentives in-game to try to combat people spamming the system and control quality of classification a bit better.

Thanks for doing science! o7