AI summary by TwinMind
RPKM, FPKM and TPM, Clearly Explained!!!
StatQuest with Josh Starmer
10m 14s
Introduction & Context
- StatQuest educational video from UNC Chapel Hill genetics department
- Focus: RNA sequencing normalization metrics (RPKM, FPKM, TPM)
- Topic specific to high-throughput RNA sequencing
- New trend: moving from RPKM/FPKM to TPM metric
RNA Sequencing Normalization Need
- Sequencing depth normalization (million part)
- Deeper sequencing runs = more reads per gene
- Must normalize to prevent bias in analysis
- Gene length normalization (kilobase part)
- Longer genes naturally get more reads than shorter genes
- Critical to remove this bias
Example Dataset Setup
- Imaginary RNA-seq dataset: 3 replicates (Rep 1, 2, 3)
- Genome: 4 genes (A, B, C, D)
- Gene lengths: A=1000bp, B=2000bp, C=3000bp, D=4000bp
- Raw read counts provided for each replicate/gene
- Replicate 3 has higher sequencing depth
- Gene B (2x longer than A) consistently gets 2x more reads
RPKM Calculation Process
- Step 1: Normalize for read depth
- Calculate total reads per replicate
- Scale by 10 (instead of 1M for simplicity)
- Create per-million scaling factors
- Divide gene read counts by scaling factors
- Step 2: Normalize for gene length
- Divide by gene length (kilobase)
- Final result: reads per kilobase per million (RPKM)
RPKM vs FPKM Distinction
- RPKM: Reads Per Kilobase Million
- Used for single-end RNA sequencing
- One read per fragment
- FPKM: Fragments Per Kilobase Million
- Used for paired-end RNA sequencing
- Accounts for fragments (not individual reads)
- Prevents double-counting when both ends of fragment map
- Sometimes only one end maps due to quality issues
TPM Calculation Process
- Key difference: Order of operations switched from RPKM
- Step 1: Normalize for gene length first
- Divide read counts by gene length
- Result: RPK (reads per kilobase)
- Step 2: Normalize for sequencing depth
- Sum RPK values for each replicate
- Create scaling factors (divide by 10 in example)
- Divide RPK values by scaling factors
- Final result: Transcripts Per Million (TPM)
Critical Difference: TPM vs RPKM Comparison
- RPKM limitation: Different total sums across samples
- Each sample has different total normalized read count
- Makes cross-sample comparisons difficult
- TPM advantage: Same total sum across all samples
- All samples sum to same value (10 in example)
- Enables direct proportion comparisons
Pie Chart Analogy
- TPM: All pies same size (10)
- Each slice represents proportion of total reads
- Easy to compare gene proportions across samples
- Example: Gene A slice larger in Rep1 vs Rep3
- RPKM: Different sized pies
- Difficult to compare slice proportions
- Cannot easily determine relative gene expression
TPM Benefits & Adoption
- Clear interpretation: Numbers directly show proportion of reads mapping to each gene
- Better for RNA-seq: Field focuses on comparing relative proportions
- More appropriate metric: Industry moving toward TPM adoption
- Same normalizations: Corrects for gene length and sequencing depth like RPKM/FPKM
