Back

Increasing information gain in animal research by improving statistical model accuracy

Waterkamp, C.; von Kortzfleisch, V. T.; Neu, C.; Kitanovski, S.; Richter, S. H.; Hoffmann, D.

2022-06-05 bioinformatics
10.1101/2022.06.03.494726 bioRxiv
Show abstract

Reduction of the numbers of laboratory animals is one of the three pillars of ethical animal research. Equivalently, information gain per animal should be maximized. A road towards this goal that is barely taken in current animal research is the more accurate statistical modeling of experiments. Here we show for a typical experiment ("open field test") with outcomes that are non-normally distributed count data, how this can be implemented and what information gain is achieved. We contrast the state of the art - the use of confidence intervals based on null-hypothesis significance testing (NHST) -, with a Bayesian approach with the same underlying normal model, and a Bayesian approach with a more accurate negative binomial model. We find that the more accurate model leads to a marked improvement of knowledge gained with the experiment, especially for small sample sizes. As experimental data that violate assumptions of simple, conventional models are frequent, our findings have wider implications.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.