🧬 biology

Machine Learning for Gene-Expression-Based Prediction of Alcohol Use Disorder with an Integrate Web-Based Dashboard

This study presents a machine learning pipeline using Elastic Net logistic regression on postmortem brain gene-expression data to predict Alcohol Use Disorder with high accuracy, accompanied by the development of the GENO web-based dashboard to facilitate interpretation and usability for laboratory researchers.

Original authors: Hanan Alotaibi, Alaa Bafail, Dania Aljeaid, Doaa Sinnari, Hebah Albatati, Mona Alkhozae, Rania M. Alhazmi, Joud Alharabi, Farah Najjar, Shahad Almasabi, Aisha Alsobhi

Published 2026-10-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hanan Alotaibi, Alaa Bafail, Dania Aljeaid, Doaa Sinnari, Hebah Albatati, Mona Alkhozae, Rania M. Alhazmi, Joud Alharabi, Farah Najjar, Shahad Almasabi, Aisha Alsobhi

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Alcohol use disorder is a complex condition that affects how people control their drinking, often leading to severe health and social consequences. While we know that environment and life experiences play a major role, science has long suspected that biology also contributes to why some people are more vulnerable than others. Inside the brain, cells communicate by turning specific genes on and off, a process that creates a molecular signature of how the brain is functioning. When a person has a disorder like alcohol use disorder, these genetic switches can change in ways that differ from a healthy brain. Researchers have spent years trying to read these genetic messages, hoping to find a pattern that could help identify the disorder. However, the human genome contains tens of thousands of genes, and the data they produce is messy and filled with noise. Finding a clear signal within that vast amount of information is like trying to hear a single conversation in a crowded stadium; it requires powerful tools to sort the important voices from the background noise.

A team of researchers at King Abdulaziz University in Saudi Arabia has taken a step forward in this challenge by building a new tool that combines advanced computer learning with genetic data to predict alcohol use disorder. They started with a collection of genetic data taken from the brain tissue of 114 people who had passed away, half of whom had a history of alcohol use disorder and half who did not. Because the data contained information on thousands of genes, the researchers first had to clean it and narrow it down to the most relevant pieces. They used a statistical method to identify which genes showed the strongest differences between the two groups, effectively filtering out the genetic static to focus on the clear signals. They then trained a computer model to look at these specific genes and learn how to distinguish between the two groups. The model they chose was designed to handle complex, interconnected data while remaining transparent, meaning the researchers could see exactly which genes were influencing the decision.

The result of this training was a system that could look at a new set of genetic data and estimate the likelihood that it came from someone with alcohol use disorder. When the researchers tested this system on a portion of the data it had never seen before, it correctly identified the status of the individuals 86 percent of the time. More importantly, the system did not just give a simple "yes" or "no" answer. It provided a probability score, telling the user how confident it was in its prediction, and it listed the specific genes that contributed most to that score. This level of detail is crucial for scientists, as it allows them to see the biological reasoning behind the prediction rather than treating the computer as a mysterious black box. To make this technology useful for other scientists, the team built a web-based dashboard called GENO. This online platform allows laboratory researchers to upload their own genetic data, run the analysis automatically, and receive a clear, visual report with the results and the list of contributing genes.

The researchers tested this dashboard with six laboratory technicians to ensure it was easy to use. The technicians were able to log in, upload a sample, and download a report in less than a minute on average, finding the process straightforward and intuitive. The study suggests that using machine learning to interpret gene expression is a viable way to identify patterns associated with alcohol use disorder, offering a new way to look at the biological underpinnings of the condition. However, the authors are careful to note that this tool is currently a research platform, not a medical diagnostic device for doctors to use on living patients. The data came from brain tissue after death, which means the tool reflects the molecular state of the brain at the end of life rather than predicting future behavior in a healthy person. Additionally, the system currently only accepts data in a specific file format and relies solely on genetic information, without yet incorporating other factors like a person's age, medical history, or environment.

Despite these limitations, the work represents a significant shift in how such research is conducted. Instead of stopping at a computer screen with a list of numbers, the team has created a bridge that connects complex data analysis to the daily workflow of a laboratory. By making the process accessible and the results interpretable, the system helps scientists move closer to understanding the molecular roots of addiction. The researchers plan to expand the tool in the future to include more diverse data types and to test it on larger groups of people. For now, the study stands as a proof of concept, demonstrating that with the right computational approach, it is possible to turn the overwhelming complexity of genetic data into a clear, actionable insight for researchers studying alcohol use disorder.

Technical Summary: Machine Learning for Gene-Expression-Based Prediction of Alcohol Use Disorder with an Integrated Web-Based Dashboard

Problem Statement
Alcohol Use Disorder (AUD) is a multifactorial condition involving genetic susceptibility, neurobiological adaptations, and environmental factors. While gene-expression profiling offers valuable molecular insights into AUD, the analysis of transcriptomic data presents significant computational challenges. These include high dimensionality (where the number of genes far exceeds the number of samples), correlated feature structures, and technical noise. Traditional statistical methods often struggle with these characteristics, leading to risks of overfitting and reduced model reliability. Furthermore, while machine learning (ML) approaches have been applied to substance use research, there is a notable gap in translating these predictive workflows into user-oriented, interpretable platforms that allow laboratory researchers to upload data, execute standardized analyses, and visualize results without requiring deep computational expertise.

Methodology
The study proposes GENO, a web-based research platform that integrates an interpretable machine learning pipeline for predicting AUD status from postmortem brain tissue gene-expression profiles. The methodology follows a structured workflow:

Dataset Acquisition: The system utilizes publicly available bulk RNA-sequencing data from the Gene Expression Omnibus (GSE253155), comprising 114 samples (58 AUD cases and 56 non-AUD controls).
Preprocessing and Feature Selection:
- Normalization: Gene expression values are standardized using StandardScaler to ensure comparable numerical scales.
- Feature Selection: To address high dimensionality, a consensus feature set is derived by merging results from two methods: the ANOVA F-test (to identify significant between-class variation) and Mutual Information (to capture non-linear dependencies).
- Data Splitting: The dataset is split into a training set (70%) and a hold-out test set (30%). All preprocessing steps (scaling and feature selection) are fitted exclusively on the training data to prevent information leakage.
Model Architecture: The core predictive engine is an Elastic Net Logistic Regression model. This approach combines L1 (Lasso) and L2 (Ridge) regularization:
- L1 promotes sparsity by shrinking coefficients of less informative genes to zero, effectively performing embedded feature selection.
- L2 distributes predictive weight across correlated features, enhancing coefficient stability and reducing overfitting.
- The model outputs a probability score ( $P(y=1|X)$ ), which is converted into a risk percentage.
- Interpretability: The system calculates the contribution of individual genes to the prediction by scaling gene expression values with the model's learned coefficients ( $Contribution_j = \beta_i \times x'_j$ ).
System Implementation: GENO is built as a web application using Python (Flask backend), React (frontend), and PostgreSQL (database). It supports data upload, automated processing, and the generation of structured reports containing risk scores and gene-level contributions.

Key Contributions
The paper outlines three primary contributions:

Feasibility Investigation: It demonstrates the feasibility of predicting AUD status from high-dimensional gene-expression profiles using an interpretable regularized machine learning approach.
Interpretable Modeling: It applies Elastic Net logistic regression to accommodate correlated transcriptomic features while providing model-based estimates of gene contributions, offering transparency often missing in "black-box" deep learning models.
Platform Integration: It translates the analytical workflow into GENO, a proof-of-concept web-based platform. This system bridges the gap between computational model development and laboratory application by providing a unified interface for data upload, analysis, visualization, and report generation.

Experimental Results
The model was evaluated on the held-out test set (30% of the data) using standard classification metrics:

Accuracy: 86%
Balanced Accuracy: 0.865
F1-Score: 0.865 (Macro-average)
ROC-AUC: 0.895

The model demonstrated consistent performance across both classes (AUD and non-AUD), with precision and recall scores ranging between 0.84 and 0.89 for each class. A preliminary usability assessment involving six laboratory technicians confirmed the platform's feasibility; all participants successfully completed core tasks (login, data upload, report generation) with an overall user experience rating of 10/10.

Significance and Claims
The authors position GENO as a predictive research platform rather than a clinically validated diagnostic tool or a system for prospectively predicting future AUD development. The study emphasizes that the gene-level contributions generated are predictive associations within the analyzed dataset, not evidence of causal relationships or independently validated biological biomarkers.

The significance of the work lies in its methodological and software-oriented approach to addressing the gap in addiction genomics. By integrating interpretable ML with a user-friendly dashboard, GENO supports laboratory researchers in translating transcriptomic machine learning from isolated experiments into accessible workflows. The authors acknowledge limitations, including the reliance on postmortem data, the use of a single data format (CSV), and the exclusion of non-genomic factors (demographic, clinical, behavioral). Consequently, the paper concludes that while the results are promising, future work must focus on integrating diverse data sources, expanding datasets, and enhancing system security to improve generalizability and clinical relevance.

More like this