Between therapy effect and false-positive result in animal experimentation.

Area of Science:

• Regenerative Medicine
• Pharmacology
• Biostatistics

Background:

• Reducing animal use in research is driven by cost and ethics.
• Small sample sizes in experiments increase the risk of false-positive results.
• Statistical significance is often the primary criterion for validating findings, but can be unreliable with limited data.

Purpose of the Study:

• To highlight the risks of false-positive results in experiments with small sample sizes.
• To propose and evaluate machine learning as a tool for validating treatment effects in regenerative therapy.
• To analyze wound healing data from mice treated with an epigenetic inhibitor.

Main Methods:

• Analyzed ear pinna punch wound healing in mice using pharmacological treatment with zebularine (an epigenetic inhibitor) versus vehicle controls.
• Compared eight treatment groups and eight control groups, each with six mice.
• Applied statistical tests (Mann-Whitney U-test) and machine learning algorithms (Naïve Bayes, Support Vector Machine).

Main Results:

• While zebularine showed a healing effect, statistically significant differences were found between control groups, indicating a risk of false positives.
• The Mann-Whitney U-test incorrectly suggested enhanced healing in some control groups.
• Both machine learning algorithms (Naïve Bayes and Support Vector Machine) correctly classified all control groups, demonstrating robustness.

Conclusions:

• Standard statistical methods can yield misleading results with small sample sizes in animal studies.
• Machine learning algorithms offer a more reliable approach for validating treatment outcomes in regenerative medicine research.
• Implementing machine learning can improve the accuracy and reliability of experimental findings, especially when reducing animal numbers.