Incorporating time as a third dimension in transcriptomic analysis using machine learning and explainable AI.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Transcriptomic data analysis, or 2DTA, measures RNA abundance at a single time point.
• While valuable, 2DTA has limitations including technical variability and a lack of temporal data.
• The impact of this 'temporality' on transcriptomic data interpretation remains unclear.

Purpose of the Study:

• To investigate the impact of time on transcriptomic data interpretation.
• To evaluate the utility of machine learning (ML) and explainable AI (XAI) in addressing the temporality problem in transcriptomics.

Main Methods:

• Utilized 25 publicly available transcriptomic datasets from MCF-7 cells at 12- and 48-hour time points.
• Applied three ML classifiers: decision trees (DT), random forests (RF), and XGBoost.
• Employed Shapley additive explanation (SHAP) for model interpretability and evaluated performance using MSE, MAE, and R-squared (DC).

Main Results:

• Significant differences in gene expression patterns were observed between the 12-hour and 48-hour datasets.
• XGBoost demonstrated superior performance over DT and RF, achieving an MSE of 0.00028, MAE of 0.00028, and R-squared of 0.95778.
• SHAP analysis provided insights into the decision-making processes of the ML models.

Conclusions:

• Time significantly influences transcriptomic data interpretation, suggesting a 'temporality' problem in 2DTA.
• Machine learning and explainable AI are effective tools for resolving the temporality issue in transcriptomics.
• This study highlights the potential of integrating temporal information into transcriptomic analyses for more robust biological insights.