project road map: https://github.com/miRTop/incubator/projects/2
People: Lorena, Ioannis, Marc, Arun,
Discussion topics:
- Universal synthetic sequences to use in the pilot
- mirtop / mirgff3 paper discussion
Action points:
- Ioannis to ask for quotation for oligo synthesis
People: Lorena, Marc, Carrie, Ioannis
Discussion topics:
- we need another source to confirm the synthetic data. If we saw the same noise, it is difficult to follow the same path
- other option is to have multiple replicates of two different tissue and test null hypothesis inside each group and between them to model the technical noise (we need to find money for that).
Action points:
- Ioannis to continue with the synthetic and illumina collaboration
- Lorena to find experiemental design for multiple replicates
- Look at new version of biorxiv from carrie's study
People: Lorena, Marc, Carrie
-
We discussed:
-
experimental design for isomir accuracy with NGS as posted here:https://docs.google.com/document/d/1XyKjQJ2R6qdES12uDK-5ffA43xgEHmcZaZ_6Vr6yWFM/edit
-
organize the results during february to split tasks in the future
-
Carrie mentioned she is talking to the company who provides the miRXplore reference to get them to give individual spikeins or proof of purification
-
To be done:
-
pilot and extended version of the isomiR spikeins (Lorena and Ioannis)
-
Carrie will ask about fragment analyzer and see if it can be used as check point of quality
-
Lorena will talk to the core sequencing to organize the fragment analyzer as quality control as well
People: Lorena, Phillipe, Ioannis, Arun, Marc
-
We discuss:
- iso_add should be iso_add3p and we should have iso_add5p and remove the sign of that and only add the number of nts involved in the variations:
iso_add3p:2. - examples in paper table should be consistent and show one example for each attribute
- isomiR shift in different databases corroborating the high counts of some isomirs that are mainly truncation events and shouldn't be there
- This papers started to look at isomiR consistency beside miRNAs (kind of similar what we are doing but I think we are going further) https://www.biorxiv.org/content/early/2018/10/30/445437?%3Fcollection=
- iso_add should be iso_add3p and we should have iso_add5p and remove the sign of that and only add the number of nts involved in the variations:
- Update paper according discussed points.
- Lorena has to send FASTA file of all sequences to Ioannis to get the features that will be used for the machine learning analysis once we can classify sequence into reproducible or not.
- Look into NTs involved in the truncation events.
- Ioannis may look at ERCC spikeins ends to see if we see the same.
People: Lorena, Ioannis, Phillipe, Aris
-
We discuss the following changes to the format:
- Coordinates based on genome will have the strand together with the chr:start-end(strand). Please chime in if there is any specific format is widely used.
- The coordinates will be 1-based, inclusive at the start and the end.
- truncation at 5p will be designated as +N, instead of -N. template additions at 5p as -N, instead of +N.
-
We showed how the other synthetic sequences inside the tewari samples produced similar isomirs.
-
We showed how mirge2.0 showed similar isomirs in the synthetic data discarting a pipeline specific issue.
- Lorena has to send FASTA file of all sequences to Ioannis to get the features that will be used for the machine learning analysis once we can classify sequence into reproducible or not.
- Lorena to analyze other data set with spike-ints: working on this
- Use cutoff tool from Phillipe group, discard low expressed reads
- Look into trimming events only to discuss about possible causes
- create a scheme from RNA material to results and spot all possibles places where artifacts can be created. It would be good for the paper and to create new questions.
People: Lorena, Ioannis, Marc, Phillipe, Lauro, Gianvito
- We discussed the equimolar data compared to plasma data to look at difference between distribution of what is the most expressed and what is the percentage of the rest of the sequences.
https://github.com/miRTop/incubator/blob/master/projects/tewari_equimolar/plasma_vs_equimolar.md
Ideas that came up:
- maybe too much noise from miR like let7*
- Ioannis confirmed that the equimolar sample may no be purified.
- Use cutoff tool from Phillipe group, discard low expressed reads
- equimolar from miRGe tool
- Ioannis mentioned extra sequences in the equimolar samples. One supplementary table from the paper shows which one are, so we can look at the variations of these ones.
- Lorena has to send FASTA file of all sequences to Ioannis to get the features that will be used for the machine learning analysis once we can classify sequence into reproducible or not.
- Marc to contact tewari's paper authors:
- whether they have equimolar and plasma samples or all is gone
- Is there information of the library size each lab did?
- Lorena to analyze other data set with spike-ints: working on this
- create a scheme from RNA material to results and spot all possibles places where artifacts can be created. It would be good for the paper and create new questions.
People: Lorena, Arun, Ioannis, Marc, Phillipe
- We discussed the equimolar data after being analyzed by bcbio pipeline: https://github.com/miRTop/incubator/blob/master/projects/tewari_equimolar/commonality.md
Ideas to explain this:
- non purified miRXplore sample
- illumina software generating 5' end that are not real
- trimming tool generating 3' end that are not real
- if isomiRs are errors from an unique sequence, and this happens after ligation, there should be a correlation of isomiRs and reference expression
- Lorena has to send FASTA file of all sequences to Ioannis to get the features that will be used for the machine learning analysis once we can classify sequence into reproducible or not.
- Ioannis to contact Qiagen and miRXplore company
- Marc to contact tewari's paper authors
- Lorena to analyze other data set with spike-ints
- Lorena to share the fasta files to be analized with miRGe2 (maybe?)
- Lorena to compare distribution of plasma sample and equimolar to see if they are the same
People: Lorena, Aarun, Ioannis
- We presented the variation in the output of IDR package when changing two of the main paramters. See Methods
- We presented an example of figure when visualizing the commonality among replicates and labs
- One strategy was to let Ioannis to contact IDR authors to ask for advice about how to adapt the code to our question
- Another strategy was to let Lorena find some statistician to join the collaboration to help in this particular step
- Lorena has to send FASTA file of all sequences to Ioannis to get the features that will be used for the machine learning analysis once we can classify sequence into reproducible or not.
- Aarun will update plots and divided by low/medium/high expression
We presented the data and results until now and discussed the following points:
- better analysis for variant calling: Ioannis
- chipseq-metrics-information to measure reproducibility: Ioannis will send information around
- mirnas matching the adapter or complemtary to it to measure the effect in expression due to the adapter used in the protocol
- measure the edit distance of the isomirs to all the mirnas in the database to spot problematic isomirs that we should remove
- Repeat plots separating between low/medium/high
- kmer analysis to explain difference among labs
The metric is called IDR (irreproducible discovery rate) and you can find it here:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3431496/
http://cran.r-project.org/web/packages/idr/index.htm
These are the papers for the bias:
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0126049
https://www.ncbi.nlm.nih.gov/pubmed/22647250
- what to do with isomiRs that map equally to other mature miRNAs
- How to filter False isomirs and True isomirs
Ideally:
- Machine learning to predict the real isomirs and remove the ones due to technical errors
features to look into that:
- PCR artifact
- Ligation
- Size selection
- Steps in protocols