Thoroughness¶

Thorough writing fully describes a result or method without providing too much, too little, or extraneous detail. This means describing not only the basic facts or concepts, but also providing sufficient context for a scientifically literate reader who is not intimately familiar with the subject matter to understand them. There is a balance between being descriptive and being concise that takes practice and experience to master.

Common mistakes¶

Mentioning key terms or concepts without describing what they are
Providing too little or too much detail (examples below)
Including tables or figures without sufficient parallel description in the text:
- In the Results section text: briefly describe what the figure/table contains and what the reader should observe about the figure
- In the Discussion section text: describe what the author believes the figure/table means and how that relates to the results of the overall study
- In the Caption text: provide specific details that aid the reader in understanding the figure
Vague or incomplete description of the methods in the Methods section
Vague or incomplete description of the results in the Results section
Vague or incomplete interpretation of the results in the Discussion section
Omitting citations justifying key claims in your text, especially if you say “Previous studies have shown…” etc

Tips¶

Make sure each program or method you used is mentioned and describe any key parameter choices (e.g. number of reads for a gene to be included in the study) in the Methods section
Do not include the specific commands or literal command line arguments (e.g. --index) in the text; it is sufficient to reference the names of programs/packages and the versions you used
When reading a published work, imagine trying to implement the analysis described in the methods section; do you feel there is sufficient detail included to do so?
Reread your own Methods section and ask the same question as above
Make a bulleted list of titles for each of your tables and figures, ordered in a logical progression
Write a results section that describes the key results and observations from your sequence of tables and figures so that the text flows from one result to the next; it should be clear to the reader why you conducted each experiment based on those described earlier
Reread your results text without looking at the figures; do you get an accurate understanding of the results from the text alone?
Make sure you reference each of your figures in the Discussion, connecting the results together to form an overall interpretation
The last paragraph of your Introduction should have included a statement about what question(s) your work sought to answer; after reading your results and discussion, do you feel you have answered those questions? If not, you should probably be more thorough.

Good example¶

A methods section from Mandelboum et al 2019, subdivided for clarity:

We analyzed 35 publicly available human or mouse RNA-seq datasets from GEO [30] (S1 Table). We sought datasets that were (1) published in recent years (mostly in 2017-18), (2) contained 2-4 replicate samples of each biological condition, (3) probed treatments with well-documented biological responses (e.g., TNFa) to ease functional interpretation and recognition of true calls by GSEA, and (4) collectively covered diverse biological processes.

The study design (# of samples, RNA-Seq datasets) is clearly defined, as is the rationale for sample selection.

We downloaded either raw count data files when provided by GEO or, otherwise, raw sequence fastq files (from SRA DB). In the latter, reads were aligned to the reference genome (hg19 for Hs and mm10 for Mm) using TopHat2 [31], and gene count data were generated using FeatureCounts [32]. We calculated cpm levels, and in each dataset analysis, we included only the expressed genes (defined as those whose expression was at least 1.0 cpm in all replicate samples of at least one of the biological condition probed in the dataset).

Specific sources of data are mentioned without providing too much detail (GEO or SRA DB). Specific reference genomes are mentioned (hg19, mm10), and published methods are mentioned and correctly cited (TopHat2, FeatureCounts). Appropriate detail is included for the important minimum cpm parameter (1.0 in all replicates of at least one biological condition).

Following this filtering step, gene counts were normalized using six different normalization methods: RPKM [7], RPKM followed by quantile normalization [25], TMM [8], RLE [9], RLE followed by RPKM, and UQ normalization followed by RPKM [11], all implemented in edgeR [33]. cqn and EDASeq (both available as Bioconductor packages) were applied to expression count data. Gene-expression FC was either calculated by dividing normalized expression levels (after adding 1.0 to both numerator and denominator and averaging over replicate samples in the treatment versus control comparisons) or estimated by edgeR regression model fit. Gene annotations were downloaded from GENCODE (v25 for Hs and vM10 for Mm) [34]. For genes with multiple transcripts, we took the length of the principal transcript (as defined by GENCODE’s annotation of principal and alternative splice isoforms [APPRIS] annotations [35]) or the length of the longest transcript if principal transcript is not defined for the gene. All statistical analyses were performed in R. Statistical significance of Spearman’s correlation was calculated using the cor.test function.

Concise language and appropriate citations, inclusion of relevant analysis choice details (e.g. “For genes with multiple transcripts…”) without too much implementation detail.

Average examples¶

Example 1¶

Original:

On the GEO (Gene Expression Omnibus) website, the paper has the repository with accession Number GSE64403. The data was acquired by using the SRA toolkit. The sample data GSM1570702 was downloaded using the fastq-dump --split-files. The --split-files is the argument that split the SRA file into two fastq files.

Writing is not concise, too little description of the data, and too much description of the methods.

Rewritten:

Paired end RNA-Seq data for the P0 timepoint sample (Dataset Accession GSM1570702) was downloaded using SRA toolkit from GEO (Series Accession GSE64403).

Example 2¶

Original:

The mapping process of the two give FASTQ files containing the RNA-seq paired end reads needed index files for the barcodes used in the RNA-Seq, in addition to annotation file of the reference sequence. Due to the computationally demanding nature of this process, a batch job script was coded to utilize number of tools available on the Shared Computer Cluster (SCC) (Table 5):

Bowtie2: is a package for alignment based on Barrows-Wheeler algorithm, which allows a margin of mismatch and randomization that reduces the memory footprint. And therefor it is fast and described as greedy algorithm. Bowtie is not Reads splice aware [6].

Tophat v2.1.1: is a splice junction mapper for RNA-Seq reads based on the aligner bowtie. Its job is to identify splice junctions between exons (Figure 5).

Boost v1.58.0 is C++ library for string and text processing.

Samtools V0.1.19: are tools to view, sort, merge and to get statistics on alignment files, that are kept in SAM format.

Unnecessary general description for methods section, too much implementation detail, Methods text should generally not include bullet points.