Related Experiment Video
Updated: Jun 24, 2026
07:31
Murine Drinking Models in the Development of Pharmacotherapies for Alcoholism: Drinking in the Dark and Two-bottle Choice
Published on: January 7, 2019
Evaluating machine learning algorithms at predicting developmental trajectories using sequential dataset truncation
Nathan Yu1, Steven Buyske2, Uthman Qureshi1
1Department of Genetics, Center of Alcohol & Substance Use Studies, Rutgers University, Piscataway, New Jersey, United States of America.
Plos One
|June 22, 2026
Summary
Machine learning accurately predicts adolescent alcohol consumption trajectories, identifying early risk factors for alcohol use disorder (AUD). This approach aids in understanding the biological basis of AUD by analyzing early behavioral data in mice.
Area of Science:
- Neuroscience
- Computational Biology
- Genetics
Background:
- Adolescent alcohol consumption is a significant risk factor for adult alcohol use disorder (AUD).
- Individual susceptibility to AUD varies due to complex, not fully understood factors.
- Identifying developmental patterns in adolescent alcohol intake may reveal biological underpinnings of AUD risk.
Purpose of the Study:
- To evaluate supervised machine learning (ML) algorithms for predicting voluntary alcohol consumption trajectories in adolescent mice.
- To assess ML model performance using sequentially truncated datasets representing different developmental time points.
Main Methods:
- Generated simulated datasets of adolescent mouse alcohol consumption.
- Applied sequential dataset truncation to train and evaluate twelve supervised ML algorithms.
- Used locally estimated scatterplot smoothing (LOESS) for curve fitting and comparative analysis of prediction accuracy.
Main Results:
- Six ML algorithms achieved over 98% prediction accuracy by the experiment's end and over 90% at the midpoint.
- Four additional ML algorithms demonstrated acceptable accuracy (77-96%).
- Two algorithms performed poorly, with accuracy declining as more data was introduced.
Conclusions:
- Supervised machine learning algorithms can reliably predict behavioral outcomes from early developmental data.
- This predictive capability can guide molecular and cellular research to identify biological drivers of alcohol consumption during adolescence.
- Early identification of AUD risk through ML may facilitate timely interventions before full behavioral phenotypes manifest.
