Although outright fraud and research misconduct definitely does occur in science sometimes, many “reproducibility” problems in science stem from poor experimental design. And some of this comes from lack of adequate training in how to set up experiments to avoid artifacts–and how to recognize them if they do occur. Some of the common ways artifacts can arise are from systematic errors including batch effects & edge effects. They get more attention (and thus have more resources for training against) in “high-throughput” (lots of samples analyzed rapidly) data collection fields (the various “-omics”) but they can also be a big problem in smaller-scale experiments such as those done in academic research labs around the world. So, here are a few tips for setting up your experiments.
https://thebumblingbiochemist.com/365...
“Batch effects” – differences in results (even of identical samples) run on different days, with different equipment, by different people, etc.
Seemingly small differences in temperature, humidity, technique, buffer pH, pipets, etc. can lead to inconsistent results
“Edge effects” – differences in results (even of identical samples) of samples towards the middle of a multi-well plate vs on the edges
Caused by uneven heating, etc. Much more on them here: https://bit.ly/edge_effects
Along similar lines, you can see differences running samples in the center of a gel vs the edges (try to avoid edges, where “smiles” and “frowns” can lead to uneven running due to uneven heating).
Other sources of position-dependent error can arise from things like:
– Multichannel pipets where one of the channels is “off” calibration-wise and/or the tip doesn’t like to grab well
– Poor and/or inconsistent technique (e.g. pipetting with a multichannel pipet is done at an angle, so the volumes in the channels are uneven)
Signal bleed-over from neighboring (typically high-concentration) samples: this can affect background subtraction of gel bands, etc.
Also, be careful when performing time-based assays that the timing of all samples is the same (don’t try to do too many things at once at the expense of some samples waiting longer than others).
Beware too of biases intrinsic to various techniques (yes, read up on how the techniques work!) and be consistent about the technique you use. For example, use a consistent method for determining sample concentrations. Due to different biases in each measurement technique, you might get very different results if you measure the concentration of the same protein using a Bradford assay, a BCA assay, or via UV280 absorption. Not to mention differences from using different protein standards to generate a standard curve (e.g., BSA or BCG) or even just from different curve fitting. The key is consistency and measuring concentrations of samples (in replicate) side-by-side at the same time. (I recommend saving some of each sample for re-measurement in the future if needed).
To account for these sorts of issues:
Run samples you hope to compare in the same batch as one another whenever possible
Include identical controls in each batch of samples to detect (and potentially account for) batch effects
Rotate the order and/or position of samples in plates, strip tubes, etc.
Replicate experiments on multiple days before jumping to any conclusions
Keep careful notes about sources of samples (date of protein purification, lot of reagent (if you’re really careful), etc.)
Try to save some of each of your samples in case you need to retest later and need a side-by-side comparison, you want to verify the sample concentration, etc.
Carefully evaluate your data for signs of systematic error.
If you spot differences you don’t expect to see, follow up!
Of course, this sort of care takes a lot of time, effort, and cost, and can thus be impractical for a lot of purposes. Typically, we’re more lax when first testing things out because we want to figure out how (and if) an experiment might work for our project, get preliminary results to help us set up “real” experiments , etc. without wasting a bunch of time, effort, & $. The careful note-taking is important, but we will often will include fewer (but not no) controls and replicates, etc.
And this is okay in many contexts. As long as 1) you’re not relying on this data and/or trying to publish it and 2) you’re not using precious samples.
Sometimes, scientists test out techniques in “proof of concept” and/or “scouting” experiments before trying to scale up and/or apply them to important datasets. (Speaking of scaling up, experimental results may not scale evenly (due to evaporation, volume loss on tubes, uneven heating, etc.))
Finished in comments
Информация по комментариям в разработке