QuantEval

About

QuantEval was released for two purposes: 1. a user can use the scripts in QuantEval to reproduce all the analyses in the following study (Hsieh et al.); 2. a user can follow the example in QuantEval to conduct the same analyses on his own study. For the first purpose, there are three modes in the QuantEval main program. (1) Reference Mode, (2) Contig Mode and (3) Match Mode. The first two modes read the quantification results and build a ambiguity cluster based on connected components for the reference transcripts and contig sequences. The match mode builds relations between contigs and reference transcripts. For the second purpose, the users are encouraged to follow the provided example to conduct new analyses on his own data.

Reference

Ping-Han Hsieh, Yen-Jen Oyang and Chien-Yu Chen. Effect of de novo transcriptome assembly on transcript quantification. Scientific Reports volume 9, Article number: 8304 (2019).

Requirement

• QuantEval main program:
  • Python3 (3.5.2)
  • Python packages: pandas (0.20.3), numpy (1.12.1)
• Generate figures and table:
  • R (3.3.0)
  • R pacakges: gridExtra, grid, stats, tidyverse, plyr, ggplot2, reshape2
• Utilities:
  • Bowtie2 (2.3.0), BLASTn (2.5.0), Flux Simulator (1.2.1), RSEM (1.2.31), Kallisto (0.43.0), rnaSPAdes (3.11.1), Salmon (0.8.2), Trans-ABySS (1.5.5), TransRate (1.0.3), Trinity (2.4.0)

Manual

• Run QuantEval individually:

python3 ./scripts/QuantEval.py --reference --contig --match --input input.json

The first three parameters (--reference, --contig, --match) indicate which mode to run and the input.json file specifies the input parameters for the QuantEval main program. The three modes can be run independantly, but one has to run both reference mode and contig mode before running the match mode. It is recommended to run three modes in sequential. Because the main program of QuantEval does not include a wrapper for quantification/sequence alignment/contig evaluation, which are essesntial steps for QuantEval main program, one might need to run quantification algorithms (i.e. RSEM/Kallisto/Salmon), sequence alignment (BLASTn) and contig evaluation (Transrate) by themselves in order to get similar analysis results in the reference research.

---

Below, we use an example dataset to explain how to use QuantEval. This example contains the following files:

• ref.fasta: the reference transcripts (In real applications, you will not have this file for the speices without reference transcripts)
• contig.fasta: the contigs assembled by short reads, e.g. read_1.fastq and read_2.fastq
• read_1.fastq read_2.fastq

Before running QuantEval,

• Run pairwise BLASTn for reference/contig mode:

# reference mode
blastn -db ref.fasta -query ref.fasta -outfmt 6 -evalue 1e-5 -perc_identity 95 -out ./blastn/ref.self.tsv 

# contig mode
blastn -db contig.fasta -query contig.fasta -outfmt 6 -evalue 1e-5 -perc_identity 95 -out ./blastn/contig.self.tsv

• Run BLASTn for the mapping of reference and contig sequence (match mode)

blastn -db ref.fasta -query contig.fasta -outfmt 6 -out ./blastn/contig_to_ref.tsv 

• Run quantification for reference/contig mode with default parameters:

# RSEM
rsem-prepare-reference --bowtie2 ref.fasta ./rsem/rsem.index
rsem-calculate-expression --paired-end --strandedness none --bowtie2 --time ref_read/read_1.fastq ref_read/read_2.fastq ./rsem/rsem.index ./rsem/rsem 

# Kallisto
kallisto index -i ./kallisto/kallisto.index -k 31 ref.fasta
kallisto quant -i ./kallisto/kallisto.index -o ./kallisto ref_read/read_1.fastq ref_read/read_2.fastq

# Salmon
salmon index -i ./salmon/salmon.index -t ref.fasta --type quasi -k 31
salmon quant -i ./salmon/salmon.index -l A -1 ref_read/read_1.fastq -2 ref_read/read_2.fastq -o ./salmon 

• Run TransRate for reference/contig mode:

# reference mode
transrate --assembly ref.fasta --output ./transrate/ref --left ref_read/read_1.fastq --right ref_read/read_2.fastq

# contig mode
transrate --assembly contig.fasta --output ./transrate/ref --left contig_read/read_1.fastq --right contig_read/read_2.fastq

• Example of input.json:

{
   "ref_fasta": "ref.fasta",
   "ref_blastn": "./blastn/ref.self.tsv",
   "ref_gtf": "ref.gtf",
   "ref_xprs_file": ["./answer/answer_xprs.tsv",
                     "./kallisto/ref/abundance.tsv",
                     "./rsem/ref/rsem.isoforms.results",
                     "./salmon/ref/quant.sf"],
   "ref_xprs_label": ["answer", "kallisto", "rsem", "salmon"],
   "ref_xprs_header": [true, true, true, true],
   "ref_xprs_name_col": [1, 1, 1, 1], 
   "ref_xprs_tpm_col": [2, 5, 6, 4], 
   "ref_xprs_count_col": [3, 4, 5, 5],
   "ref_transrate": "./transrate/ref/contigs.csv",
   "contig_fasta": "contig.fasta",
   "contig_blastn": "./blastn/contig.self.tsv",
   "contig_xprs_file": ["./kallisto/contig/abundance.tsv",
                        "./rsem/contig/rsem.isoforms.results",
                        "./salmon/contig/quant.sf"],
   "contig_xprs_label": ["kallisto", "rsem", "salmon"],
   "contig_xprs_header": [true, true, true],
   "contig_xprs_name_col": [1, 1, 1], 
   "contig_xprs_tpm_col": [5, 6, 4], 
   "contig_xprs_count_col": [4, 5, 5],
   "contig_transrate": "./transrate/contig/contigs.csv",
   "match_blastn": "./blastn/contig_to_ref.tsv",
   "output_dir": "./QuantEval/"
}

---

• Run example:

cd example
python3 ../scripts/QuantEval.py --reference --contig --match --input ./example.json

---

One can also import the functions in utilities.py to built their own analysis pipeline.

• Construct connected component for reference only:

from utilities import construct_sequence, filter_blastn, intersect_match, construct_grap
import copy

input_file = dict()
input_file["contig_ref_file"] = ["./answer/answer_xprs.tsv",
                                 "./kallisto/abundance.tsv",
                                 "./rsem/rsem.isoforms.results",
                                 "./salmon/quant.sf"]
input_file["ref_xprs_label"]: ["answer", "kallisto", "rsem", "salmon"],
input_file["ref_xprs_header"]: [true, true, true, true],
input_file["ref_xprs_name_col"]: [1, 1, 1, 1], 
input_file["ref_xprs_tpm_col"]: [2, 5, 6, 4], 
input_file["ref_xprs_count_col"]: [3, 4, 5, 5],   
ref_seq_dict = construct_sequence('ref.fasta')
ref_self_blastn = filter_blastn('./blastn/ref.self.tsv')
read_expression(input_file, ref_seq_dict, 'ref')
ref_self_match_dict = intersect_match(ref_self_blastn, ref_seq_dict, copy.deepcopy(ref_seq_dict))
ref_uf, ref_component_dict = construct_graph(ref_seq_dict, ref_self_match_dict)

print(ref_uf.component_label)
print(ref_uf.parent)

---

• Run all the analysis in the study (time consuming):

./pipelines/run_analysis.sh

---

• Output format

column	description
match_name	alignment (ref.contig.strand)
contig_name	target contig name
ref_name	target ref name
accuracy	accuracy of the alignment
recovery	recovery of the alignment
contig/ref_length	length of contig/ref
contig/ref_tr_transrate_score	transrate score of contig/ref
contig/ref_xprs_tpm/count_quantifier	quantification result of contig/ref
contig/ref_component	label of connected component of contig/ref
contig/ref_component_size	number of sequences in the connected component of contig/ref
contig/ref_component_contribute_xprs_tpm/count_quantifier	proportion of TPM/read count (RPEA) in the connected component of contig/ref
contig/ref_component_relative_xprs_tpm/count_quantifier	TPM/count of contig/ref / highest TPM/count in the same connected component
contig/ref_component_max_xprs_tpm/count_quantifier	highest TPM/count of contig/ref in the same connected component
contig/ref_component_avg_xprs_tpm/count_quantifier	average TPM/count of contig/ref in the same connected component
contig/ref_component_tot_xprs_tpm/count_quantifier	total TPM/count of contig/ref in the same connected component
ref_gene_contribute_xprs_tpm/count_quantifier	proportion of TPM/read count in the gene of ref
ref_gene_relative_xprs_tpm/count_quantifier	TPM/count of ref / highest TPM/count in the same gene
ref_gene_max_xprs_tpm/count_quantifier	highest TPM/count of ref in the same gene
ref_gene_avg_xprs_tpm/count_quantifier	average TPM/count of ref in the same gene
ref_gene_tot_xprs_tpm/count_quantifier	total TPM/count of ref in the same gene
length_difference	the difference of length between contig and reference
xprs_tpm/count_error_quantifier	quantificaion error for the estimated abundance of contig

Note that this is the superset of the output fields (the match mode). The content of output will be different depends on reference/contig/match mode (e.g. one can only find the columns start with contig from contig mode), but one can find all the description on the table above.