russelllab / kinaseresistance Goto Github PK

https://stackoverflow.com/questions/37121767/svg-with-clickable-links-in-shiny-not-clickable

re:svg, I was looking this up for my shiny application today and people seem to recommend incorporating svgs via "".
It did not work for RShiny (I found another way to do it there) but otherwise seemed to be a promising approach.

https://agrussell.slack.com/archives/C04P305PK61/p1683191322269059
Point 6, can you introduce another parameter in the main function that tells how to sort the genes in the alignment?
z.B.

• value 1 = rank by similarity (you could write a BLAST protocol to do this?)
• value 2 = rank by known information quantity (default, as of now)
• value 3 = rank by p-sites (you can use the kinase_project DB of PostgreSQL on peveolution2)

At the front end, I will create a drop-down menu, from which the user selects the type of sort, and then I call the same main function.

I think you can use the PostgreSQL DB to fetch the fasta sequences and PTM information. You can connect to the DB in your code:

import psycopg2

def connection():
    '''Function to connect to postgresql database'''
    mydb = psycopg2.connect(
                            database = "kinase_project",
                            user = "gurdeep",
                            password = "hellokitty",
                            host = "localhost",
                            port = "5432")
    return mydb

mydb = connection()
mydb.autocommit = True
mycursor = mydb.cursor()

and from the command-line on pevolution2: psql -U gurdeep kinase_project

Let me know if I can provide or do something else or if you have a better alternative in mind. :-)

Collect all possible features and create an FM (feature matrix) to be used for ML

• Fetch sequences from UniProt
• Map Pfam positions to the sequences
• Draw a bubble plot/heatmap

1. "
   EGFR
   P00533/V689M
   P00533/E1005R
   P00533/D1006K

MAP2K3
P46734/S218E
P46734/T222E

Actually, multiple newlines cause a crash I think."

2. "(gives "You called GET") and putting in various things just crashed (e.g. FGFR2/C278F)"

@JCGonzS

Other little things for the WebApp (some of these are repetitions, but people under 40 apparently do not ever read anything that isn't in a tweet - according to some podcast I just heard - so...)

• name of the kinase and the Uniprot ID would be nice in the first info box. We could even do synonyms.
• #41
• hyperlinks wherever possible
• fix the "known mutation" labels to remove boomerie punctutation, ECO1234 and add hyperlinks to the PubMed links.

we want sections about prior info below the name:

• in this kinase at this position
• #40
• in other kinases at this position
• in others kinases +/- 2 residues or maybe +/- 5 residues
• perhaps give the user the option to rank kinases in the alignment in different ways? Rank by sequence similarity or rank by information quantity or rank by n-phosphosites, etc. #37 (comment)
• help pages need to be there. (#39 (comment))
• Change the URL to activark or whatever
• (re: 5 I'm not sure if c should be before b but it probably doesn't really matter)
• labels regarding which region of the kinase canonical structure we are in (P-loop, A-loop whatever)

• Silent variants
• Without slash
• Incorrect format

z.B. BRAF/V600E is missing

Add active links for p-sits and known variants. Remove "ECO" etc from the info and just have the text and the link.

@tschmenger

refer point 2
https://agrussell.slack.com/archives/C04P305PK61/p1682938480543719

Let's go with colour-blind-friendly shades.

The top one (red/orange) is deactivating
Third from the top (bluish) is resistance
the bottom-most one (greenish) is activating

I can fx this in my scripts and can you do this in yours?

• Currently, the sequences are sorted based on identity
• Take the top 25, 50% and 75% and regenerate the scores
• The idea is that perhaps by limiting the homology distance, we can get better resolution

#41 is also fixed. Follow the comments below to update the alignment

1. Take an input kinase and find out the top 30 kinases based on seq identity
2. Do an hmmalign of all the candidates and prepare an output aln
3. Make dictionary with mutations and region annotations

1. Normalize the data correctly. For example, homolog scores should be done around -1 to 1 and not 0 to -1 (maybe)
2. Add structure features

• make a function that created output folder and call it every time
• make a dedicated function that stores meta data in a dic, which is returned
• display basic information at top of results page
• connect summary to results page
• in AJAXChart, check if output already exists before running the predictor again

Write a short tutorial like in precog(x)

1. Take all the human kinases
2. Align them first using hmmalign to the Pkinase profile
3. Then look at the alignment with ADR mutations with the help of Torsten's tool
4. Fix it manually, and then build an HMM out of it.

Create an alignment based on what is around the given position of interest or by selecting kinases that have the most information.

Error handling for both webApp and command-line z.B. if gene or ACC does not exist, if the position is outside domain

Write tests for both webApp and command-line

@gurdeep330 told me that Rob
"wants the ADR heatmap to NOT change i.e. it should be done for the entire alignment (>400 kinases) even if the user is looking at top 10 or 20, say"

The most efficient solution I could think of immediately would be to do the counts for the 'heatmap' separately and read it from a dictionary (using alignment positions rather than protein positions I think) rather than going through each letter of the whole alignment to sum up the values, each time we recalculate/redraw the alignment.

Draw a GRID violin plot to describe the role of each feature i.e. show multiple features in the same plot.

• If the given position is a PTM site
• if the given position in an ADR site
• 1 and 2 if known at other positions
• if 1 and 2 in neighbouring residues

Say in the info box if the position is upstream or downstream or the same position. Use some labels in HTML? Same for variants too.

Use embeddings from Uniprot as features

https://agrussell.slack.com/archives/C04P305PK61/p1682938727615919

• Show aln postitions and not hmm positions
• Check that predictions for outside the kinase domain are not made
• Check if changing to "Activating" helps in the search
• Add +1/-1 in the aln/hmmpos columns of the summary table

• Download buttons
• Color scores
• "known ADR"

@tschmenger
So Rob suggested having the mouseover information in the Mechismo format z.B.
kinase/WTpos; mt_type
EGFR/T790; R
T is the WT AA and R is the mutation type