Lag penalized weighted correlation for time series clustering.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Clustering biological time series requires similarity measures tailored to temporal structures.
• High-throughput assays generate sequential data like gene expression over time.
• Existing methods may not capture synchronized temporal patterns effectively.

Purpose of the Study:

• To introduce a new similarity measure for time series clustering.
• To group time series exhibiting similar temporal behaviors despite potential timing discrepancies.
• To enhance the interpretability of clusters in biological time series data.

Main Methods:

• Proposed Lag Penalized Weighted Correlation (LPWC) as a novel clustering similarity measure.
• LPWC aligns time series profiles and down-weights based on introduced temporal lags.
• Evaluated LPWC against existing time series and general clustering algorithms.

Main Results:

• LPWC successfully recovered true clusters in a simulated biological dataset.
• The method demonstrated advantages over existing clustering algorithms.
• LPWC identified distinct temporal patterns in yeast and axolotl regeneration data.

Conclusions:

• LPWC effectively groups time series with correlated temporal changes, accommodating asynchronous patterns.
• The method penalizes large temporal shifts, preserving pattern integrity.
• An R package for LPWC is publicly available for use in biological data analysis.