Overview #
Computational prediction tools estimate the potential functional impact of genetic variants by integrating evolutionary, structural, and biochemical information.
While some predictions are directly used for ACMG classification (e.g., PP3/BP4 via REVEL, CADD, and SpliceAI), SeqSMART also visualizes a broad set of additional predictors to give users a comprehensive view of variant behavior across different analytical dimensions.
In SeqSMART, computational predictions are displayed using interactive visual panels designed to help expert users interpret complex data intuitively.
1. Primary Predictors Used for ACMG Classification #
SeqSMART automatically incorporates three high-confidence predictors directly into PP3/BP4 evidence scoring:
| Predictor | Purpose | Integration |
| REVEL | Meta-predictor combining multiple algorithms for missense impact. | Used to quantify PP3/BP4 based on ClinGen-defined thresholds. |
| CADD (Combined Annotation Dependent Depletion) | Estimates deleteriousness across all variant types using integrated annotations. | Used when REVEL is unavailable or for non-missense variants. |
| SpliceAI | Deep-learning predictor of splicing alterations. | Used to assess PP3/BP4 or BP7 when splicing disruption is relevant. |
Each of these predictors is directly incorporated into SeqSMART’s ACMG rule engine and also visualized interactively for user reference.
2. Visualization in SeqSMART #
SeqSMART presents computational prediction data through four interactive visual components, each providing a different layer of insight into variant behavior:
A. REVEL and CADD Score Panel #
- Purpose: To visualize deleteriousness scores from two widely used meta-predictors.
- Display: A dual horizontal bar chart —
- Pink bar: REVEL score
- Blue bar: CADD score
- Pink bar: REVEL score
- Interpretation:
- High REVEL (>0.77) or CADD (>25) values suggest likely pathogenicity.
- Low values indicate tolerance or benignity.
- High REVEL (>0.77) or CADD (>25) values suggest likely pathogenicity.
- Usage Tip: These scores are directly used for ACMG PP3/BP4 logic but can also be cross-compared visually for model consistency.
B. SpliceAI Radar Chart #
- Purpose: To visualize predicted splicing alterations.
- Display: A circular radar plot showing probabilities of:
- Acceptor Loss (blue)
- Donor Loss (yellow)
- Acceptor Gain (green)
- Donor Gain (red)
- Acceptor Loss (blue)
- Interpretation:
- Scores >0.5 suggest likely splice alteration (supporting PP3).
- Lower scores (<0.2) typically indicate preserved splicing (supporting BP4/BP7).
- Scores >0.5 suggest likely splice alteration (supporting PP3).
- Note: SeqSMART uses these same predictions in its automated ACMG criteria but also displays them for transparency.
C. Conservation Prediction Chart #
- Purpose: To summarize evolutionary conservation across species — an important clue for variant significance.
- Display: A 9-axis radar graph showing normalized conservation scores from widely used predictors:
- gerp_s – GERP++ score (phylogenetic conservation)
- pri_ph_cons – Primate phastCons
- mam_ph_cons – Mammalian phastCons
- ver_ph_cons – Vertebrate phastCons
- pri_phylo_p – Primate phyloP
- mam_phylo_p – Mammalian phyloP
- ver_phylo_p – Vertebrate phyloP
- zoo_pri_phylop – Primate phylop zoo track
- zoo_ver_phylop – Vertebrate phylop zoo track
- gerp_s – GERP++ score (phylogenetic conservation)
- Interpretation:
A broad, high-filled area (toward the outer edge) indicates strong cross-species conservation, implying the variant occurs in an evolutionarily important region.
Conversely, small or low profiles suggest evolutionary tolerance.
D. Missense Predictor Panel #
- Purpose: To provide additional algorithmic opinions beyond REVEL, offering broader interpretive context.
- Display: A four-quadrant visualization presenting outputs from:
- SIFT – Predicts deleteriousness based on amino acid conservation.
- PolyPhen-2 – Evaluates structural and functional impact on proteins.
- PROVEAN – Predicts functional effects of amino acid substitutions or indels.
- MutPred2 – Assesses molecular mechanism alterations (if available).
- SIFT – Predicts deleteriousness based on amino acid conservation.
- Interpretation:
Each predictor’s classification (e.g., Benign, Deleterious, Neutral) and score are displayed for user comparison.
These results are not used for ACMG scoring, but provide valuable context to complement the REVEL and CADD data.
3. How to Use These Visualizations #
SeqSMART’s computational prediction dashboard allows users to:
- Compare multiple prediction models side-by-side.
- Identify consistent signals across independent predictors.
- Quickly detect discordance between models (e.g., REVEL benign but CADD moderate).
- Support manual ACMG interpretation by validating or questioning automated results.
These visualization tools are designed for expert interpretation, providing transparency while maintaining analytical rigor.
4. Notes for Users #
- Computational predictions should never be used in isolation for final classification.
- Always interpret predictions in conjunction with:
- Population frequency (gnomAD)
- Functional studies (PS3/BS3)
- Phenotypic consistency (PP4)
- Gene mechanism (LoF relevance)
- Population frequency (gnomAD)
- SeqSMART ensures that each predictor is updated regularly and normalized for cross-comparison accuracy.
SeqSMART integrates and visualizes a wide spectrum of computational predictions — from meta-predictors and splicing models to conservation and structural tools.
By combining interpretive automation (for PP3/BP4) with transparent visual exploration, SeqSMART empowers genomic professionals to make data-driven, evidence-balanced variant interpretations.
SeqSMART Predictive Insight Principle:
See the numbers. Understand the biology.