Chapter 7: Step 5 — Functional Annotation Engines & In Silico Predictors
Tested Stacks:
GATK 4.6.1
Samtools 1.21
Nextflow 24.10
Seurat v5.1
Python 3.12
Prerequisites: Basic Linux CLI & Sequencing Formats
📚 Chapter 7 of 10
0% Course Completed
Chapter 7: Step 5 — Functional Annotation Engines & In Silico Predictors
Once the raw VCF file is generated, the clinical interpretation bridge officially begins. A raw VCF line contains only physical coordinates; annotation engines query genome transcript models to determine the biological effect on RNA and protein structures, while machine learning predictors estimate evolutionary and functional damage.
Step 5 Technical Specifications
- Input File: Raw or filtered VCF file (
.vcf.gz). - Tools Used: Ensembl VEP (Variant Effect Predictor), ANNOVAR, SnpEff.
- Scoring Predictors: CADD, REVEL, AlphaMissense, SIFT, PolyPhen-2, SpliceAI.
- Output File: Annotated VCF / TSV file with full transcript, protein, and damage metrics.
- Clinical Utility: Standardizes transcript selection via MANE Select, maps codons to amino acid alterations, and calculates in silico pathogenicity scores required for ACMG criterion PP3 and BP4.
1. Transcript Selection Standards: The MANE Project
In the past, diagnostic laboratories frequently experienced catastrophic disagreements because one lab annotated variants against an Ensembl transcript while another used a RefSeq transcript with different exon numbers. The MANE (Matched Annotation from NCBI and EMBL-EBI) initiative resolved this by establishing a single universal transcript for every human protein-coding gene:
- MANE Select: Exactly one high-confidence transcript per gene that matches 100% identically between RefSeq (NM_) and Ensembl (ENST_) across all exons, splice junctions, and protein products. Always use MANE Select as the clinical reporting baseline.
- MANE Plus Clinical: Supplementary transcripts specifically curated for genes where pathogenic variants occur in alternative exons not captured in the Select transcript (e.g. SCN5A, KCNQ1).
2. Consequence Prediction Taxonomy
| Consequence Class | Molecular Mechanism | SO Term | ACMG Clinical Impact |
|---|---|---|---|
| Nonsense / Stop Gained | Codon converted to stop (TAA, TAG, TGA), triggering Nonsense-Mediated Decay (NMD) | stop_gained | PVS1 Candidate (Very Strong Pathogenic) if LoF is mechanism |
| Frameshift Indel | Insertion or deletion not a multiple of 3, disrupting triplet reading frame | frameshift_variant | PVS1 Candidate (Very Strong Pathogenic) |
| Canonical Splice Site | Mutation disrupting conserved ±1,2 dinucleotides (GT donor or AG acceptor) | splice_donor/acceptor_variant | PVS1 Candidate (causes exon skipping / intron retention) |
| Missense / Non-Synonymous | Single nucleotide alteration swapping one amino acid for another | missense_variant | Evaluated via REVEL, AlphaMissense, PM1, PM5, PP2, PP3 |
| Synonymous / Silent | Nucleotide change preserves identical amino acid due to genetic code degeneracy | synonymous_variant | BP7 Candidate (Supporting Benign) if splice sites unaffected |
3. In Silico Pathogenicity Scoring Engines
Modern clinical guidelines require combining multiple independent algorithmic predictors rather than relying on a single score:
- CADD (Combined Annotation Dependent Depletion): An integrative framework that scores the deleteriousness of single nucleotide variants and insertions/deletions. CADD scores are Phred-scaled:
CADD ≥ 20represents the top 1% most deleterious substitutions in the human genome;CADD ≥ 30represents the top 0.1%. - REVEL (Rare Exome Variant Ensemble Learner): The gold-standard ensemble predictor for missense variants recommended by ClinGen. REVEL integrates 13 individual tools (MutPred, FATHMM, VEST, PolyPhen, SIFT, PROVEAN). Calibrated thresholds:
REVEL ≥ 0.75$ ightarrow$ Triggers ACMG criterion PP3 (In silico damaging).REVEL ≤ 0.15$ ightarrow$ Triggers ACMG criterion BP4 (In silico benign).
- AlphaMissense (Google DeepMind): Built upon AlphaFold 3 protein structure representations and language models. Evaluates amino acid substitutions in their true 3D spatial conformation, achieving unprecedented accuracy in uncharacterized genes.
- SpliceAI (Illumina): A 32-layer deep residual neural network predicting non-canonical cryptic splice donor and acceptor gain/loss up to 10,000 bp into introns. Score $\ge 0.50$ indicates high probability of splice disruption.
# Comprehensive Functional Annotation using Ensembl VEP
vep -i patient_germline.raw.vcf.gz -o patient_annotated.vcf.gz --vcf --compress_output bgzip --assembly GRCh38 --species homo_sapiens --cache --dir_cache /opt/vep_cache --everything --mane_select --plugin CADD,/opt/cadd/whole_genome_SNVs.tsv.gz --plugin REVEL,/opt/revel/revel_grch38.tsv.gz --plugin AlphaMissense,file=/opt/alphamissense/AlphaMissense_hg38.tsv.gz --plugin SpliceAI,snv=/opt/spliceai/spliceai_scores.raw.snv.hg38.vcf.gz --fork 16
⚠️
Common Pitfalls & Troubleshooting Checklist
- Java Heap Exhaustion (GATK / Picard): If processes terminate with
java.lang.OutOfMemoryError, set your environment variable explicitly:export _JAVA_OPTIONS="-Xmx16g -Djava.io.tmpdir=/tmp". - Index Timestamp Mismatch: If tools complain that a BAM index is missing or out of date, regenerate with
samtools index -@ 4 sample.bam. - Conda Channel Conflicts: Avoid ABI incompatibilities between bioconda and conda-forge by setting strict priority:
conda config --set channel_priority strict. - Chromosome Naming Contamination: Always verify whether reference FASTA headers use
chr1or1; divergent naming causes silent alignment drops.
Was this guide helpful to your research?
Help us maintain 100% accurate, peer-reviewed bioinformatics standards.