Overview #
Functional evidence provides some of the most direct insights into a variant’s biological effect.
Within the ACMG/AMP framework, PS3 and BS3 represent strong criteria based on the results of in vitro or in vivo functional studies.
However, because such data are often limited to a small subset of variants, SeqSMART combines AI-assisted literature mining, automated data extraction, and expert curation to identify, score, and validate functional evidence efficiently and transparently.
1. What Are PS3 and BS3? #
| Criterion | Description | Evidence Direction |
| PS3 (Pathogenic Strong 3) | “Well-established in vitro or in vivo functional studies supportive of a damaging effect on the gene or gene product.” | Supports pathogenicity when functional studies demonstrate a clear loss or alteration of function caused by the variant. |
| BS3 (Benign Strong 3) | “Well-established in vitro or in vivo functional studies show no damaging effect on protein function or splicing.” | Supports benignity when validated assays show normal function or no molecular impact. |
Both criteria rely on reproducible and validated experimental data, typically derived from biochemical assays, splicing analyses, cellular localization studies, or animal models.
Because of their demanding evidence requirements, PS3 and BS3 are powerful but infrequently applicable in practice.
2. SeqSMART’s Approach to Functional Evidence #
To make functional evidence more accessible and scalable, SeqSMART uses a hybrid AI–human curation framework consisting of three main stages: automated evidence discovery, relevance scoring, and expert validation.
2.1 Automated Data Extraction (NLP Engine) #
SeqSMART employs an advanced Natural Language Processing (NLP) model to systematically extract functional evidence from multiple data sources:
- ClinVar functional evidence fields
- Scientific publications from PubMed (titles, abstracts, and full texts)
- Supplementary materials, figures, or captions describing assay outcomes
The NLP model identifies key experimental details — assay type, biological system, phenotype observed, and functional outcome — and links them to the variant record.
2.2 Evidence Scoring and Relevance Detection #
Each extracted reference undergoes contextual scoring to assess its reliability and strength:
- Assay Type: e.g., enzymatic activity, protein localization, splicing analysis, or cell viability.
- Result Quality: clear, reproducible results versus inconclusive or conflicting findings.
- Reproducibility: consistency across independent studies or citations.
SeqSMART’s internal scoring model generates a functional evidence profile summarizing all relevant experimental results associated with each variant.
2.3 Expert Curation #
To maintain scientific integrity, all NLP-derived evidence is reviewed by human curators or qualified experts before affecting ACMG scoring.
Experts evaluate whether each study:
- Meets ACMG’s definition of a “well-established” functional assay.
- Directly assesses the molecular consequence of the variant.
- Demonstrates a clearly damaging (PS3) or non-damaging (BS3) effect.
- Is relevant to the gene’s biological mechanism and disease context.
Only after expert confirmation does the system activate PS3 or BS3 for classification.
3. Manual Adjustment by Users #
Because high-quality functional data remain rare, PS3 and BS3 are often not automatically assigned by default.
However, SeqSMART gives expert users full control to adjust these criteria manually:
- If you have internal experimental results (e.g., in-house splicing assays, enzymatic studies), you may apply PS3 or BS3 manually in the interface.
- You can also review and validate all evidence collected by the NLP system in the Functional Evidence section of the variant page.
- Each manual adjustment is logged with author name, timestamp, and justification to preserve transparency.
This flexibility enables institutions to incorporate unpublished or internal functional data into their variant assessment workflows.
4. Transparency and Traceability #
SeqSMART ensures full transparency in functional evidence evaluation:
- All NLP-extracted studies, sentences, and PubMed references are visible directly within the variant interface under the PS3/BS3 section.
- The interface highlights whether evidence was automatically detected, curator-validated, or manually added.
- The source of each record (ClinVar, literature, internal upload) is clearly indicated for reproducibility.
5. Continuous Improvement #
SeqSMART’s functional evidence module evolves over time:
- Ongoing machine learning retraining improves NLP precision and recall for functional text extraction.
- The evidence library expands as new studies become available.
- Version control ensures that all PS3/BS3 assignments are linked to the software version and model configuration used at the time of analysis.
As the SeqSMART evidence base grows, the number of variants with automated PS3/BS3 assessment will continue to increase, supporting more robust and efficient interpretation.
Summary #
SeqSMART transforms the traditionally manual process of evaluating functional evidence into a hybrid AI-assisted workflow.
By combining automated literature extraction, contextual evidence scoring, and expert validation, SeqSMART ensures that PS3 and BS3 are applied with rigor, transparency, and flexibility.
SeqSMART Functional Evidence Principle:
Automate discovery — validate by expertise — classify with confidence.