Babesia Time Course — Reproducible Analysis

This repository provides a minimal, reviewer-ready pipeline to reproduce:

1. Noise removal / batch correction (edgeR TMM; outlier clusters removed)
2. LOESS smoothing → one curve per gene (average technical → average biological)
3. PCA on cleaned counts (optional FC ≥ 1.5 filter; off by default)

The code is RStudio-friendly and runnable from the command line.

---

Repository Structure

• input/ # <- Input data
• scripts/ # 01_noise_removal.R, 02_loess.R, 03_pca.R
• tables/ # outputs: cleaned tables, LOESS tables, PCA scores
• figures/ # outputs: PCA plot
• legacy/ # archived older scripts (not used by the pipeline)

Requirements

• R (≥ 4.1 recommended)
• Install these packages once:

1. openxlsx
2. edgeR
3. stringr
4. ggrepel
4. readr
5. tidyverse

How to Run (Command Line )

In a terminal:

git clone https://github.com/umbibio/Babesia_time_course.git

From the repository root (Babesia_time_course):

STEP 1: Noise Removal / Batch Correction

Rscript scripts/01_noise_removal.R

Outputs

• tables/raw_counts_normal_growth_clean.xlsx — cleaned counts (first col GeneName)

• tables/experiment_design_clean.xlsx — cleaned design with Batch (clusters 3 & 4 removed)

Summary

• Filters low expression (CPM > 2 in ≥ 3 samples)

• edgeR TMM normalization → logCPM

• Hierarchical clustering on samples → cut into 4 clusters → drop clusters 3 & 4

• Standardizes gene column to GeneName

STEP 2: LOESS Smoothing (One Curve per Gene)

Rscript scripts/02_loess.R

Outputs

• tables/BdAllData.xlsx with columns:

• Gene, maxTime, minTime, maxValue, minValue, FoldChange

• BdyValues1..50 (smoothed curve values)

• BdNormyValues1..50 (same curve, normalized to sum = 100)

• If some genes had < 3 usable timepoints after cleaning:

Summary

• Averages technical replicates within each biological replicate (BE1+BE2 → BE, CK1+CK2 → CK) per timepoint, using available columns

• Averages biological replicates (BE vs CK) per timepoint, using available values

• Fits one curve per gene with adaptive LOESS (more stable with few points);

• falls back to linear interpolation when needed

• Evaluates each curve at 50 points; keeps legacy column names for compatibility

• tables/loess_skipped_genes.tsv (gene ID, reason, times used)

STEP 3: PCA on Cleaned Counts

Rscript scripts/03_pca.R

Outputs

• tables/pca_scores.tsv — PC1/PC2 for each timepoint (after averaging reps)

• figures/pca_scores.pdf — publication-ready PCA plot

Summary

• edgeR TMM → logCPM on cleaned counts

• Averages all replicates at each timepoint (BE/CK; 1/2)

• Optional amplitude filter across time (set min_fold_change <- 1.5 in the script to enable)

• PCA on timepoints (rows) × genes (columns)

Running in RStudio (Optional)

• Open the repo in RStudio.

• Source & run, in order:

• scripts/01_noise_removal.R

• scripts/02_loess.R

• scripts/03_pca.R

• Inspect outputs under tables/ and figures/.

Naming Conventions & Assumptions

• Sample names: BdC9-<BE|CK>- (e.g., BdC9-BE1-6)

• Time is parsed from the trailing integer (hours).

• LOESS uses only available timepoints per gene after cleaning.

• A minimum of 3 usable timepoints is required to fit a curve.

• Seeds are set where applicable for reproducibility.

License

This repository is licensed under the MIT License.