ajw2329/junctioncounts

Utils for identifying pairwise alternative splicing events and quantifying them using junction reads.

junctioncounts's Issues

Assess GLM performance

Assess performance of GLMs on simulated data. TP would be a 'ground truth' dPSI of the same sign as the junctionCounts calculation given a p-value lower than chosen alpha in the case of comparison across levels of a single factor.

In the case of more complicated hypotheses such as testing interaction terms, a TP might be the presence of a 'ground truth' ddPSI of the same sign as junctionCounts calculation between pairs of A levels across B levels (e.g in the case of full: PSI ~ A + B + A:B vs red: PSI ~ A + B

Add weights for PSI_lo, PSI_hi

Add weights (sum of included, excluded counts) for PSI_lo and PSI_hi values. This will allow the use of either in binomial GLMs.

Add general dPSI calculation script

Generalize the dPSI code in the benchmarking Rmd files to R scripts with CLI.

Take advantage of bam fetching

Paired-end processing is already set up to use fetching. This could be taken advantage of to only parse reads that overlap events.

TO DO:
Take splice_lib_events.bed as input and parse - while looping through the intervals, pass the coords to bam.fetch() in read_pair_generator. Modify SE read processing to also use bam.fetch().

For bootstrapping, make use of idxstats to get the total number of reads. The all_read_info_list will then contain the subset of informative reads with n 0s for n uninformative reads.

Missing RI events

E.g. ZSWIM8 (chr10:75561074-75561606) from A1 hg19 merged.gtf. This gets generated internally but ends up filtered for some reason.

Update README.md

Need comprehensive docstrings

Not only for this but also (and more importantly) for splice_lib.

Implement distance constraint for AP, AT events

Currently alternative polyadenylation/alt-TSS events are specified by two exons with the same inner coordinate (splice site) but different outer coordinates. The outer coordinates can differ by as little as 1 nt. This most likely allows for many false positives, particularly when input GTFs come from transcriptome assemblers.

python packaging

Turn junctionCounts into a proper installable python package.

assign_reads_unstranded uses erroneous 3-argument calls to append_matching

junctionCounts version 0.1.0

running with option --forward_read unstranded

The following exception Traceback was generated:

File "<PATH TO junctionCounts_0.1.0>/junctionCoutns.py" line 611 in assign_reads_unstranded
TypeError: append_matching() takes exactly 4 arguments (3 given)

It appears that the code was updated in "assign_reads_stranded", which has 2 calls with 4 arguments of this form:

append_matching( candidate_isoforms, matching_events, event_junction_dict, junction)
append_matching( candidate_isoforms, matching_events, event_eij_dict, eij)

But the code in "assign_reads_unstranded" has 2 calls with the old erroneous 3-argument form:

append_matching( matching_events, event_junction_dict, junction)
append_matching( matching_events, event_eij_dict, eij)

Hopefully the associated changes to make this work are not too burdensome.

Sol Katzman
UC Santa Cruz Genomics Institute