sheynkman-lab / biosurfer Goto Github PK

example:

Features to implement

• skinny bars for UTRs, thick bars for CDSs
• mutations as lollipops (red=disease, green=benign, orange=variant of unknown significance)
• show expression levels from short-read RNA-seq data

Need a class that represents disease mutations (genetic variants, mendelian disease mutations, complex disease variants, cancer autosomal and somatic variants) that are mapped to Position and Residue objects.

Placeholder for module -> isomodules/isomut.py

Code that could be used as a start -> isomodules_in_progress/mutation*.py

Write code so that creation of each gene object can be done separately (use parallel python)

Same parallelization can be applied to the alignment of objects.

Need a word that represents new exon sequence that harbors a start codon...

pop up
spliced in
????

Currently - the isoimage code only takes in same-strand ORFs for plotting.
We need to think about what to with antisense transcripts, which are prevalent in our long-read data, as well as GENCODE.

Need to discuss how to describe the cut-out splice

proposed hierarchy:

• Gene owns Transcripts
• Transcript owns ORF(s)
• ORF owns UTRs (5' and 3')
• ORF and UTR point to constituent Exons and chain of Positions
• ORF also points to constituent CDSs and chain of Residues
• Exon owns chain of Positions
• CDS owns chain of Residues

Jared idea - to rework the underlying code for representing features as ranges. He thinks this could allow for “lighter-weight” representation of objects. Potentially an optimization project after the “user interface” base code is working.

How was “requirements.txt” populated?

Thinking about language for the translational differences. If we say "frameshift", that is expected to mean a ribosomal "slip", stochastic or programmed, that causes a shift in the frame of translation for the same transcript.

I have not explicitly found a term that describes differences in relative frame of translation between different isoforms. I can potentially ask the directory of GENCODE about this.
We may need to say, “usage of a different translational frame”

https://en.wikipedia.org/wiki/Ribosomal_frameshift

Jared found a reference to an "exitron"

Should we name it that?

https://en.wikipedia.org/wiki/Exitron

We can consult with some experts that study the process.

Exitrons also can have frame-preserving or frame-shifted structures.

Need a word for this situation:

might try implementing this as sorting a list of tuples; if that doesn't work out, copy the optimized old code

mutations as lollipops (red=disease, green=benign, orange=variant of unknown significance)

Essentially make Biomolecule and its subclasses "wrappers" for interval objects (see example)

What is the best way to structure many different groupings between features, of different modes and sizes, and at different scales?

Starting a thread here for ideas on how to encode group-specific information efficiently:

At the time of this post, making queries from iso-objects in biosurfer requires chained attribute accesses. For example, align_groups[2].frmf.blocks[1].first.res.doms might pull out the set of domains to which the first residue in a frameshifted region is mapped.

Would it be possible to let users make SQL-style queries? What are the pros and cons in terms of ease of code development, performance, user accessibility, etc?

Things that could make it easier for users and/or the annotation algorithm to identify splicing events that affect protein sequence:

• have Residue objects that correspond to stop codons
• replace EmptyResidue objects with positions (0-length ranges) within ORFs
• explicitly show the presence of introns in full string representation of alignment

Idea to think about - should we have an "annnotation" class which will be a flexible container to hold annotation information.

Annotation information examples:

• GO term for a gene
• More information about a domain
• A paper or series of studies that show functionality for a particular isoform

Figure out how to add a table to the labels

To show data like this:
ETV2.pdf

show expression levels from short-read RNA-seq data

Possibility - class which represents the abundance of the isoform (or isoform sub-element, such as a junction or exon) in human tissues, cell lines, or disease-relevant samples.

Original code (now deleted) input the GTEx data and had transcript abundances across ~30 human tissues (see below).

This class may be helpful for comparing abundances of events (e.g., exon skipping) versus whole-isoforms. It may be helpful in comparing short-read-based versus long-read-based expression values.

It could also be used to plot expression visualizations in the isoform imager module. For example - Farilie's Swan program has an example.

    def __init__(self, expr_dict):
        self.expr_dict = expr_dict # tissue -> rpkm
        self.avg_expr = self.compute_avg_expr()

    def __getitem__(self, tiss):
        # when expr_obj fetch by key (tissue), return value
        return self.expr_dict[tiss]

    def compute_avg_expr(self):
        tot = 0
        for k, v in self.expr_dict.items():
            v = float(v)
            tot += v
        avg_expr = tot/len(self.expr_dict)
        return avg_expr```

Think about how multiple features connected to the same ORF should be represented, so that one can readily know which features (type, and number) exist for each ORF/exon/residue

Come up with a “name” for this pattern, frameshift-then-shift-back-into-register

Are there naming conventions for split-codons associated with AAs that span two exons?

We could consult with Alain Laeder. or Jason Underwood about this.

A way to visualize isoforms interactively? Like a google-map scheme? Can zoom in and out, automatically squeeze introns for whatever group of isoforms are being displayed. Jared suggests Bokeh or Plotly.

Isocreatefeat.create_and_map_domain -> if run more than once, redundant domains created

Alignments affected:

• ARHGEF1-203|204
• ICAM4-202|201
• OSCAR-204|206

Occurs at exon junctions.

• Strand
• Nucleobase (ACGT)
• AminoAcid (rename object class to "Residue"?)
• Frame
• Alignment block category

Implement subclasses for features - by downloading data and coding them in
types of features:
this is the name of the attr linked to the iso_obj:
(for example, orf.dom retrieves current_dom from orf.doms)

• dom (cat can be dbd, reg, act, repr, other) - domain
• binding residue, activate site
• lm - linear motif
• idr - intrinsically disordered region
• sstruc - secondary structure
• ptm (cat can be phospho, acetyl, etc.) - post-translational modification
• cons - conservation
• frm - translational frame
• isr (cat can be constitutive, subset, or isoform-specific) - isoform-specific region
  Related to this is “Fractional splice” - Fractional splice code
  For PTMs, look at Phosphosite+, Cell Signalling Technologies

• skinny bars for UTRs, thick bars for CDSs

I was looking at the code in which the Y axis label is set. I was wondering how the physical Width of the Y axis label is determined? If you submit a longer string will the left-hand side of the canvas expand in response?

code to reproduce:

gene = <gene object>
aln_grps = isocreatealign.create_and_map_splice_based_align_obj([[gene.repr_orf, orf] for orf in gene.other_orfs])
for aln_grp in aln_grps:
    isocreatefeat.create_and_map_frame_objects(aln_grp)

known examples:

• GCDH
• TIMM50

Should we have a single module that holds utility functions or have several separate modules?

Currently - distance between tracks is 1. Need to find a way to dynamically determine the spacing of tracks (e.g., with domains and lollipop figures, one isoform may be taller than another)

How long does it take on average to instantiate a full gene object? Versus load from a pickle?

Will probably need to fix this while addressing #21.

Show the 5’UTR and 3’UTR as thin bars
E.g., UTR is light blue and thinner; CDS i darker blue and thicker

This can only happen after Transcript class is implemented (#21).

need to find examples

Legend for any shading/hatching/picture elements
Hatching - need a legend for the translational frame

Show a green bar for first methionine; and red bar for the stop codon

Need to indicate the Appris principles (e.g., asterisk next to transcript_name, and indicate what asterisk means in the legend)

Include a statement (near bottom left or right is usually good) saying:
*GENCODE APPRIS principle isoform