Time-Series Analysis of Periodically Expressed Genes in the Yeast Cell Cycle

Overview

Written as an AP Biology research report, this study turns the textbook account of cell-cycle gene regulation into a quantitative question: can periodically expressed genes be identified from data alone?

Research Questions

• Can genes whose expression changes periodically through the cell cycle be identified from data?
• When do those genes reach peak expression, and how does that timing relate to their biological function?

Data

• GEO dataset GSE4987, a Saccharomyces cerevisiae cell-cycle microarray time series
• Cells were arrested in G1 with alpha factor and sampled at 5-minute intervals while kept synchronized
• Time-series expression values for about 8,000 genes

Method

• Loaded and preprocessed the data in Python with pandas and NumPy
• Removed genes with missing values and applied per-gene z-score normalization to focus on the shape of each expression pattern
• Fast Fourier Transform: ranked genes by the power of the first positive frequency component as a measure of periodicity
• Detected expression peaks with SciPy's find_peaks and used the standard deviation of peak intervals to assess regularity
• Visualized the results with Matplotlib and seaborn: per-gene time series, z-score heatmaps, clustered heatmaps, and peak-time bar charts

Finding periodic genes

1. 1

   GSE4987

   Yeast synchronized in G1 with alpha factor, sampled every 5 minutes

2. 2

   Clean and normalize

   Drop genes with missing values, z-score each gene

3. 3

   FFT ranking

   Score periodicity by the power of the first positive frequency

4. 4

   Peak regularity

   find_peaks and the standard deviation of peak intervals

5. 5

   Visualize

   Time series, z-score and clustered heatmaps, peak-time bars

Results

• Selected the 10 most periodic genes and examined the top 5 in detail
• The heatmap showed peaks shifting diagonally across genes, revealing staggered activation over time, and clustering grouped genes with similar patterns
• Peak expression of the top genes concentrated around the G2 and M phases, consistent with the preparation for cell division

Follow-up — Statistical Re-verification and Prediction

• Applied Fisher's g-test with multiple-testing correction to four cell-synchronization datasets
• Of 724 previously listed cell-cycle genes, only 55 were confirmed in two or more datasets, showing the gap between visual selection and statistical significance

Statistical re-verification across four synchronization datasets

• Previously listed cell-cycle genes724
• Confirmed in two or more datasets55

View data

Item	Value
Previously listed cell-cycle genes	724
Confirmed in two or more datasets	55

• Built logistic regression and random forest models that predict cell-cycle association from expression patterns alone
• The random forest's cross-validated performance was above random guessing but limited, likely because experimental conditions and noise structures differ across datasets

Limitations

• Microarray data carry technical noise, and genes with low expression or missing values are hard to analyze
• FFT captures the strength of an oscillation but not the exact timing or phase of its peaks
• Future work: single-cell RNA-seq analysis to capture differences between individual cells