Big picture:
- Work on relevant topics
- Get advice from many sources on relevance
- Avoid topics that may be quickly outdated
- Search Google/PubMed for previous work on topic
- Will the topic have an impact on biomedical research?
- Does the topic need a new method?
- Seek out excellent collaborators / labs
- Care about the computational and quantitative aspects
- Care about technical bias and experimental design
- Lots of luck involved
- Not every research product will get any attention
- Diversify and see what works
Details:
- Work on small subsets of data first
- Do lots of spot checks early on
- 1 day of spot checks saves months of erroneous analyses
- With large data, don’t immediately run on all data, but try a small subset (something that will finish in < 10 min)
- Check results by eye
- Visualizing overall patterns (boxplots, PCA, MA, heatmap, dendrograms)
- Visualize individual examples (data for one gene/feature)
- Visualize lower-level patterns (genomic coverage of reads)
- Successful software requires more than a good algorithm
- Documentation and user support take time but go a long way
- When possible, provide online HTML resources, gets more eyeballs and is more search friendly than PDF
Writing:
- Don’t bury the most important message
- Put the important message in title, and in abstract
- Most of the people who see your paper will only read the title, then some fraction will read the abstract. Only a minority will actually open the paper, and most of those will look at Figures and the headings in the Results section
- Use simple language, repeat same words/structure for simplicity