aTIS Constraint Analysis

Evolutionary constraint analysis of alternative translation initiation site (aTIS) regions using gnomAD variants.

Overview

This analysis compares constraint metrics between:

• aTIS regions: 5'UTR extensions and N-terminal truncations
• Canonical CDS: Standard protein-coding sequences

Following the approach from Whiffin et al. 2024 (Genome Biology), we use:

1. Simple paired comparisons (aTIS vs canonical within same gene)
2. LOEUF stratification to test correlation with gene-level constraint

Dataset

Source: SwissIsoform MANE isoform results

• 4,757 aTIS features (2,963 extensions + 1,794 truncations)
• gnomAD v4.1 variant counts and constraint metrics (GRCh38)

Only essential columns retained:

• Identifiers: gene_name, transcript_id, feature_id, feature_type
• Feature info: feature_start, feature_end, feature_length_aa
• gnomAD variant counts: missense, synonymous, nonsense, frameshift

Methods

See METHODS.md for detailed methodology.

Key Approach

Constraint metrics calculated for both aTIS and canonical regions:

1. O/E Ratios (Observed/Expected)

   • Missense, Synonymous, LoF O/E
   • aTIS expected counts scaled from gnomAD canonical using length ratio: exp_aTIS = exp_canonical × (AA_aTIS / AA_canonical)
   • O/E < 1 = constrained (fewer variants than expected)
   • O/E > 1 = tolerant (more variants than expected)

2. Variant Densities (variants per amino acid)

   • Missense, Synonymous, LoF densities

3. Paired Comparisons (within-gene)

   • Delta: aTIS - canonical
   • Ratio: aTIS / canonical

Data Integration

• Transcript-based merge: SwissIsoform features matched to gnomAD v4.1 constraint metrics via Ensembl transcript IDs (version-stripped)
• Quality filters: Requires gnomAD match, LOEUF score, valid lengths, complete constraint metrics
• Result: 4,757 features from 2,593 genes across 2,950 transcripts with paired aTIS/canonical metrics

Pipeline

# 1. Prepare data (download gnomAD v4.1, merge with SwissIsoform features)
python src/01_prepare_data.py
# Output: data/merged_features.csv (4,757 features, 29 columns)

# 2. Calculate constraint metrics (O/E ratios, densities, paired comparisons)
python src/02_calculate_metrics.py
# Output: data/features_with_metrics.csv (4,757 features with all metrics)

# 3. Visualize distributions (exploratory plots)
python src/03_plot_metrics.py
# Output: results/figures/{all,extensions,truncations}/

Status: Exploratory analysis complete. Statistical testing (step 04) pending.

Preliminary Observations (Exploratory Plots)

From visual inspection of distributions (results/figures/):

1. aTIS regions show higher O/E ratios than canonical CDS

   • aTIS O/E distributions shifted right (higher O/E = less constrained)
   • Most aTIS regions have O/E > 1 (more variants than expected)

2. Paired comparisons show consistent patterns

   • Median delta O/E > 0 for most features (aTIS less constrained)
   • Small fraction (~10-15%) show negative delta (aTIS more constrained than canonical)

3. Extensions vs Truncations

   • Both feature types show similar constraint patterns
   • Separate plots enable visual comparison

Note: These are descriptive observations. Formal statistical testing (paired tests, LOEUF stratification) pending in step 04.

Outputs

data/
├── gnomad/
│   └── gnomad.v4.1.constraint_metrics.tsv  # Auto-downloaded (92MB)
├── merged_features.csv          # SwissIsoform + gnomAD v4.1 merged (4,757 features, 29 columns)
└── features_with_metrics.csv    # All calculated constraint metrics (4,757 features)

results/figures/
├── all/                         # Plots for all 4,757 features
│   ├── 01_oe_distributions.png
│   ├── 02_density_distributions.png
│   ├── 03_oe_scatter.png
│   └── 04_delta_distributions.png
├── extensions/                  # Plots for 2,963 extensions
│   └── [same 4 plots]
└── truncations/                 # Plots for 1,794 truncations
    └── [same 4 plots]

References

• Methodology: Inspired by Whiffin et al. 2024. "Differences in 5'untranslated regions highlight the importance of translational regulation of dosage sensitive genes." Genome Biology 25:111. DOI: 10.1186/s13059-024-03248-0

• Data sources:

  • Chen S, Francioli LC, Goodrich JK, et al. (2024) A genome-wide mutational constraint map quantified from variation in 76,156 human genomes. bioRxiv. [gnomAD v4.1, GRCh38]
  • SwissIsoform MANE alternative isoform database

• Detailed methods: See METHODS.md