{ "topic": { "id": 405237, "title": "Information Accretion explainer ", "authorName": "Clara De Paolis", "commentCount": 18, "votes": 68, "postDate": "2023-04-26T16:49:08.702000" }, "comments": [ { "id": 2293265, "authorName": "Nilay K. Bhatnagar", "votes": 1, "postDate": "2023-06-09T04:21:35.850000", "content": "

Thank you, This was a really helpful read

" }, { "id": 2238279, "authorName": "Vishal Joshi", "votes": 1, "postDate": "2023-04-28T11:15:24.063000", "content": "

The paper talks in section 3.3 Measuring the quality of function prediction

\n
\n

As the decision threshold is moved from its minimum to its maximum value, the pairs of forumla.gif will result in a curve in 2D space

\n
\n

What is min and max value of threshold?

" }, { "id": 2244243, "authorName": "Clara De Paolis", "votes": 0, "postDate": "2023-05-03T14:31:16.747000", "content": "

Think of it like a traditional precision-recall or ROC curve. tau is the decision threshold for predicting a positive term (class). To make a PR or ROC curve, you would compute precision and recall at many threshold (tau) values between 0 and 1 to populate every point in the curve. It's the same idea here but the curve axes are remaining uncertainty and misinformation.

" }, { "id": 2244448, "authorName": "Vishal Joshi", "votes": 0, "postDate": "2023-05-03T16:25:21.013000", "content": "

Thanks @claradepaolis for the replies!

" }, { "id": 2238271, "authorName": "Vishal Joshi", "votes": 1, "postDate": "2023-04-28T11:06:27.317000", "content": "

Hi, This is great explanation as I have been trying to implement this paper in Java.
\nWhen you form the set of parent of terms Pa(v) for a given term v, do you go all the way to the root of ontology or just parent terms of v and not beyond that?

\n

E.g. v has parents p1 and p2.
\np1 has parent p11
\np11 has parent \"root\"
\np2 has parent p21
\np21 has parent \"root\"

\n

So, which one of the following is true?
\nPa(v) = {p1, p2, p11, p21, root}
\nor
\nPa(v) = {p1, p2}

" }, { "id": 2244227, "authorName": "Clara De Paolis", "votes": 1, "postDate": "2023-05-03T14:20:50.020000", "content": "

The parents are the nodes that are directly connected to node v, i.e. Pa(v) = {p1, p2} in your example.

" }, { "id": 2236346, "authorName": "Guilherme Trajano", "votes": 1, "postDate": "2023-04-26T19:20:02.103000", "content": "

Great explanation!

" }, { "id": 2382037, "authorName": "JKC", "votes": 0, "postDate": "2023-08-09T14:39:46.477000", "content": "

Thanks for your explanation. It is very helpful for me.

" }, { "id": 2379530, "authorName": "AbaoJiang", "votes": 0, "postDate": "2023-08-08T07:47:43.190000", "content": "

Hi, sorry for the late question.
\nWill information accretion be recalculated in the final evaluation phase? Thanks

" }, { "id": 2330151, "authorName": "KagglePro", "votes": 0, "postDate": "2023-07-04T18:44:00.380000", "content": "

That's really precise and intuitive, thank you!

" }, { "id": 2243066, "authorName": "Vishal Joshi", "votes": 0, "postDate": "2023-05-02T17:04:22.467000", "content": "

What is the intuition behind negative information accretion (ia)? It happens when the descendant term has is associated with more number of proteins than its parent.

" }, { "id": 2243076, "authorName": "Norix", "votes": 1, "postDate": "2023-05-02T17:10:10.633000", "content": "
\n

What is the intuition behind negative information accretion (ia)? It happens when the descendant term has is associated with more number of proteins than its parent.

\n
\n

That sounds like an annotation error. Per definition, a protein with GO term X should always have all of X's parents, too.
\nHowever, keep in mind that this only works for is_a and part_of relations, not has_part or regulates.
\nhttp://geneontology.org/docs/ontology-relations/

" }, { "id": 2244221, "authorName": "Clara De Paolis", "votes": 3, "postDate": "2023-05-03T14:18:33.747000", "content": "

@norixmb is correct. In addition, this can happen if you are using a different structure (ontology) than the one that generated (propagated) the original terms. If the structure has changed, the parents of nodes may have changed. The training terms and obo file we provide should match and produce no negative values of IA.

" }, { "id": 2239143, "authorName": "Ericka42", "votes": 0, "postDate": "2023-04-29T07:54:48.013000", "content": "

That was really helpful!Thank you for sharing!

" }, { "id": 2237504, "authorName": "Mohammed Hamdy", "votes": 1, "postDate": "2023-04-27T17:13:47.063000", "content": "

Thanks, that was really helpful.

" }, { "id": 2236708, "authorName": "Ujjwal Grover", "votes": 1, "postDate": "2023-04-27T04:33:22.973000", "content": "

Really helpful. Thanks 🙏

" }, { "id": 2389462, "authorName": "Hunter Wrynn", "votes": 0, "postDate": "2023-08-14T04:43:02.263000", "content": "

Thanks, that was really helpful.

" }, { "id": 2308884, "authorName": "Svyatoslav Filatov", "votes": 0, "postDate": "2023-06-19T09:41:37.127000", "content": "

neat explanation, thanks!

" } ], "index": { "id": "405237", "title": "Information Accretion explainer ", "authorName": "", "commentCount": "18", "votes": "68", "postDate": "2023-04-26 16:49:08.702000" }, "competition": "cafa-5-protein-function-prediction" }