Simulating single cell RNA library generation (scRNA-seq)
This repository is as part of the Uni Basel course <E3: Programming for Life Science – 43513>. To test the accuracy of scRNA-seq data we generated the synthetic data. That is, we reconstruct the properties of the experimental data set and determine whether the computational analysis can recover properties of the data that was assumed in the simulation. This is never trivial since setting the ground truth is much needed in the computational method to evaluate the result.
Synopsis
As part of the sub-project, we implemented python code for selecting terminal fragments. Detailed distribution used for the selecting fragments can be found below, summarised in this paper.
Next Generation Sequencing (NGS) technology is based on cutting DNA into small fragments and their massive parallel sequencing. The multiple overlapping segments termed “reads” are assembled into a contiguous sequence. To reduce sequencing errors, every genome region should be sequenced several dozen times. This sequencing approach is based on the assumption that genomic DNA breaks are random and sequence-independent. However, previously we showed that for the sonicated restriction DNA fragments the rates of double-stranded breaks depend on the nucleotide sequence. In this work we analyzed genomic reads from NGS data and discovered that fragmentation methods based on the action of the hydrodynamic forces on DNA, produce similar bias. Consideration of this non-random DNA fragmentation may allow one to unravel what factors and to what extent influence the non-uniform coverage of various genomic regions.
As a whole project, we implemented a procedure for sampling reads from mRNA sequences, incorporating a few sources of “noise”. These include the presence of multiple transcript isoforms from a given gene, some that are incompletely spliced, stochastic binding of primers to RNA fragments and stochastic sampling of DNA fragments for sequencing. We will then use standard methods to estimate gene expression from the simulated data. We will repeat the process multiple times, each time corresponding to a single cell. We will then compare the estimates obtained from the simulated cells with the gene expression values assumed in the simulation. We will also try to explore which steps in the sample preparation have the largest impact on the accuracy of gene expression estimates.
Usage
CLI arguments:
fasta (required): Path to FASTA file with cDNA sequences
counts (required): Path to CSV/TSV file with sequence counts
output (required): Output file path
mean: Mean fragment length (default: 300)
std: Standard deviation fragment length (default: 60)
size: Chunk size for batch processing (default: 10000)
"""Takes as input FASTA file of cDNA sequences, a CSV/TSV with sequence counts, and mean and std. dev. of fragment lengths. Outputs most terminal fragment (within desired length range) for each sequence."""
Output:
Text file with most terminal fragments for each sequence.
To install package, run
pip install .
Development
To build Docker image, run
docker build -t terminal_fragment_selector .
License
MIT license, Copyright (c) 2021 Zavolan Lab, Biozentrum, University of Basel
Given a set of full-length DNAs, apply fragmentation and select end fragments within a certain length range. Each copy of a transcript is fragmented independently of the others.
Input:
Fasta-formatted file of transcript sequences
Csv-formatted ("TranscriptID,GeneID,Count) with transcript copy numbe
Mean fragment lengths
Standard deviation of fragment lengths
Output:
Fasta-formatted file with terminal fragments from the input transcripts that fall within the desired range of length (mean +/- 2 standard deviations).
First, a number of break points should be chosen so that the expected fragment length is the one provided (input 3). Then, the location of these points in the transcripts should be sampled. Finally, if the terminal fragment of the respective transcript falls within 2 std.dev. (input 4) from the provided mean (input 3), it is saved in the output file. The process should be repeated independently for each copy (input 2) of each transcript (input 1).
Given a list of cDNA sequences and their counts, fragments are created from these sequences by cutting them in n random locations, and the most terminal (within desired length range) fragment for each sequence is stored.
Explicitly, we take as input:
FASTA file with cDNA sequences
CSV file with sequence counts (associated by Seq ID)
Mean and standard deviation of fragment lengths
The script then iterates through each transcript copy, randomly samples n cut locations (n = len(transcript)/mean(fragment length)), generating n+1 fragments. The most terminal fragment within the desired length range (mean +/- std. deviation (input 3)) is saved. Once all sequences have been processed, output is a FASTA-formatted file containing these terminal fragments.
Can possibly use a non-uniform random distribution when selecting cutting points as this would be more biologically/technologically realistic? As Mihaela stated, IRL these fragments are run through a gel and the ones within a band range are taken. These are likely not to be uniformly distributed in size.
README description
Terminal fragment selecting
Simulating single cell RNA library generation (scRNA-seq)
This repository is as part of the Uni Basel course <E3: Programming for Life Science – 43513>. To test the accuracy of scRNA-seq data we generated the synthetic data. That is, we reconstruct the properties of the experimental data set and determine whether the computational analysis can recover properties of the data that was assumed in the simulation. This is never trivial since setting the ground truth is much needed in the computational method to evaluate the result.
Synopsis
As part of the sub-project, we implemented python code for selecting terminal fragments. Detailed distribution used for the selecting fragments can be found below, summarised in this paper.
As a whole project, we implemented a procedure for sampling reads from mRNA sequences, incorporating a few sources of “noise”. These include the presence of multiple transcript isoforms from a given gene, some that are incompletely spliced, stochastic binding of primers to RNA fragments and stochastic sampling of DNA fragments for sequencing. We will then use standard methods to estimate gene expression from the simulated data. We will repeat the process multiple times, each time corresponding to a single cell. We will then compare the estimates obtained from the simulated cells with the gene expression values assumed in the simulation. We will also try to explore which steps in the sample preparation have the largest impact on the accuracy of gene expression estimates.
Usage
CLI arguments:
fasta (required): Path to FASTA file with cDNA sequences
counts (required): Path to CSV/TSV file with sequence counts
output (required): Output file path
mean: Mean fragment length (default: 300)
std: Standard deviation fragment length (default: 60)
size: Chunk size for batch processing (default: 10000)
sep: Sequence counts file separator (default: ",")
"""Takes as input FASTA file of cDNA sequences, a CSV/TSV with sequence counts, and mean and std. dev. of fragment lengths. Outputs most terminal fragment (within desired length range) for each sequence."""
Output:
To install package, run
Development
To build Docker image, run
License
MIT license, Copyright (c) 2021 Zavolan Lab, Biozentrum, University of Basel
Contact
zavolab-biozentrum@unibas.ch
Original issue description
https://git.scicore.unibas.ch/zavolan_group/pipelines/scrna-seq-simulation/-/issues/6
Terminal fragment selection
Given a set of full-length DNAs, apply fragmentation and select end fragments within a certain length range. Each copy of a transcript is fragmented independently of the others.
Input:
Output: Fasta-formatted file with terminal fragments from the input transcripts that fall within the desired range of length (mean +/- 2 standard deviations).
First, a number of break points should be chosen so that the expected fragment length is the one provided (input 3). Then, the location of these points in the transcripts should be sampled. Finally, if the terminal fragment of the respective transcript falls within 2 std.dev. (input 4) from the provided mean (input 3), it is saved in the output file. The process should be repeated independently for each copy (input 2) of each transcript (input 1).
Pipeline overview description
https://git.scicore.unibas.ch/zavolan_group/pipelines/scrna-seq-simulation The resulting cDNAs are fragmented according to the parameters specified by input #7, and the end-fragments of the cDNA are selected.
Project design plan
https://git.scicore.unibas.ch/zavolan_group/tools/terminal-fragment-selector/-/issues/1
Given a list of cDNA sequences and their counts, fragments are created from these sequences by cutting them in n random locations, and the most terminal (within desired length range) fragment for each sequence is stored.
Explicitly, we take as input:
The script then iterates through each transcript copy, randomly samples n cut locations (n = len(transcript)/mean(fragment length)), generating n+1 fragments. The most terminal fragment within the desired length range (mean +/- std. deviation (input 3)) is saved. Once all sequences have been processed, output is a FASTA-formatted file containing these terminal fragments.
Can possibly use a non-uniform random distribution when selecting cutting points as this would be more biologically/technologically realistic? As Mihaela stated, IRL these fragments are run through a gel and the ones within a band range are taken. These are likely not to be uniformly distributed in size.