Disorder predictions for all reviewed UniProt proteins with "membrane" subcellular localization

IDRs in membrane proteins

About

The writeup below walks through how to answer the question "what fraction of membrane proteins have IDRs".

The tl/dr: on average 55% of membrane proteins [as defined here] have IDRs.

Data

This directory (WITH the associated sequences), which are not included on GitHub, are deposited on Zenodo.

Steps

Install software

Install metapredict

pip install metapredict

Done.

Find the data

Using the "Advanced Search", I downloaded all reviewed proteins on UniProt tagged as having a "membrane" subcellular localization.

There are, of course, much more detailed ways one could get these data (e.g., requiring transmembrane helices, integral membrane protein filtering etc, etc), but I figured for purposes here, I'd just do this...

Download the data

From this, I downloaded all the FASTA sequences. This triggers the following file to be downloaded:

uniprotkb_cc_scl_term_SL_0162_AND_revie_2024_11_09.fasta

Preprocessing the data

Next I wanted to fix any sequences with non-standard amino acids by converting those residues to "standard" amino acids. I did this using pfasta, a commandline tool that's part of protfasta which comes with metapredict.

Did this simply with

pfasta --invalid-sequence convert-all -o membrane_proteins.fasta uniprotkb_cc_scl_term_SL_0162_AND_revie_2024_11_08.fasta

The --invalid-sequence flag tells pfasta what to with an invalid sequence, and -o tells it what the ouput file should be called.

Running this (~5 seconds) generates the following output:

........................
pfasta version 0.1.13+0.gc1786bd.dirty
Please report bugs to:
https://github.com/holehouse-lab/protfasta
........................

[INFO]: Reading in the file uniprotkb_cc_scl_term_SL_0162_AND_revie_2024_11_08.fasta
[INFO]: Read in file with 1004330 lines
[INFO]: Parsed file to recover 117691 sequences
[INFO]: Converted 697 sequences to valid sequences
[INFO]: Writing new sequence file [membrane_proteins.fasta]...

Predicting all IDRs

Run one of metapredict's command-line utilities to predict all IDRs in these proteins.

metapredict-predict-idrs membrane_proteins.fasta -o membrane_protein_idrs.fasta

This takes ~20 seconds on my desktop with an A4500 GPU or ~5 min with my Macbook Pro. We're then left with a file called membrane_protein_idrs.fasta. This is a FASTA file that the header format:

> IDR_START= IDR_END=

e.g.

>sp|B4R1D8|HH_DROSI Protein hedgehog OS=Drosophila simulans OX=7240 GN=hh PE=3 SV=1 IDR_START=0 IDR_END=94
MDNHSSVPWASAASVTCLSLDAKCHSSSSSCSSKSTASSISASPETQTMRHIAHTQRCLS
RLTSLVALLLIVLPMMFSPAHSCGPGRGLGRHRA

Hacky way to get number of proteins with IDRs

Get the number of unique proteins in the main file:

cat membrane_proteins.fasta | grep ">" | awk -F "|" {' print $2 '} | sort | uniq |wc

Which returns

117691 117691 829609

So we have 117,691 unique proteins (NB: we could also just have gotten this without the awk -F "|" {' print $2 '} | sort | uniq part BUT we do need that for the number of unique proteins with IDRs so good to be consistent.

cat membrane_protein_idrs.fasta | grep ">" | awk -F "|" {' print $2 '} | sort | uniq |wc

Which returns

64971 64971 458829

So on average 55% of membrane proteins [as defined here] have IDRs.

Bonus!

In addition we can predict all disorder scores rather than discrete IDRs using

metapredict-predict-disorder membrane_proteins.fasta -o membrane_protein_disorder_scores.csv

This takes ~5 min on my Macbook Pro. The data for these are included here for completeness/convenience.