-
Notifications
You must be signed in to change notification settings - Fork 0
/
proposal.txt
68 lines (37 loc) · 2.71 KB
/
proposal.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Predict Protein Structure using Machine Lerning techniques
file:///Users/chris61015/Downloads/ijms-17-02118.pdf
https://github.com/alevchuk/kmercount-classifier
http://www.tandfonline.com/doi/abs/10.1080/08927020701377070?src=recsys&journalCode=gmos20
https://cs224d.stanford.edu/reports/LeeNguyen.pdf
https://arxiv.org/pdf/1603.06430.pdf
http://file.scirp.org/pdf/JBiSE_2016042713533805.pdf
http://msb.embopress.org/content/12/7/878
http://biorxiv.org/content/early/2016/09/16/073239
Deep Learning => Classification of sequence of protein
=> find sub-classification
=> What parameters could we use?
1. Sequence Database: Evolutionary => easier
2. Scoring functions
3. Classification => “Classification of secondary structure”
4. K-means
Clustering on Environmental Variable=>String match
Ramachandran
http://www2.warwick.ac.uk/fac/sci/moac/people/students/peter_cock/python/ramachandran/other/
http://www.proteinstructures.com/Structure/Structure/Ramachandran-plot.html
https://www.quora.com/What-is-a-Ramachandran-plot-How-do-you-read-one-and-what-information-can-you-learn-from-one
http://foldit.wikia.com/wiki/Secondary_structure_prediction_tools
Ramachandran Plot:
Why should we do clustering? Isn’t is a deterministic model? Or is it OK to do clustering?
Find all psi, phi means knowing the structure?
Step 1. Find several molecules with similar structure(How to find structure/Is random ok?)
Step 2. Divide psi degrees and phi degrees into sub slot respectively
Step 3. Count the occurence probability of a slot
Step 4. Use a probability model to depict joint distribution
Step 5. Use a ML algorithm(Soft-margin SVM) to generalize prob
Step 6. Using test dataset to validate data
Step 7. Explain the result
Q1. When drawing a Ramachandran-plot for an unknown protein, is it likely that we could get all phi-psi but know nothing about whether it belongs to alpha or beta? Or we will get phi-psi and its label (belong to alpha, beta, etc) altogether?
Q2. Can we use CATH to find a bunch of proteins with similar structure? Are proteins with similar structure share similar Ramachandran-plot? Or we could randomly choose any protein do conduct our classification model?
Q3. Our results could be slots which are assigned with probabilities that whether a specific psi-phi angle belongs to alpha or beta. In this case, how could we take advantage of it to help us classify unknown proteins? Or we do not need to do that?
Q4. I have heard that you’ve mentioned about clustering the Rakmachandran-Plot. Clearly I know how to cluster data, but I wasn’t clear how could we take the advantage of it. I mean, if I find a best K in Kmeans to cluster a Rakmachandran-Plot for a protein, what should I do next?
Thanks for your help.