Multiple Myeloma Digital Twin
Expert-grade therapy response prediction using CrisPRO.ai genome-scale language model. We quantify mutation disruption in critical cancer pathways (RAS/MAPK and TP53) to predict patient sensitivity vs resistance — with live, transcript-aware scoring and strict data hygiene.
Why It Matters
- Reduce VUS from ~40% to ~15% (target) to unblock decisions and experiments.
- Explainable therapy ranking with citations speeds tumor board alignment.
- Provenance (run IDs) ensures repeatability and auditability.
What We Delivered
- Variant Insight chips; Therapy Fit table; Pathway View; Trials shortlist.
- Toxicity Risk chip; CRISPR Readiness (demo).
- Exports (JSON/CSV) and provenance across views.
Validation Metrics
Rigorous validation across multiple datasets demonstrates our platform's accuracy and reliability
ClinVar AUROC (total n=53,210)
Overall accuracy across coding/non-coding SNVs and non-SNVs
SpliceVarDB AUROC (n=4,950)
Exonic/intronic splice prediction accuracy (~82.5–82.6%)
BRCA1 Supervised (coding SNV)
AUROC 0.94, AUPRC 0.84 — oncology benchmark
Research Findings
Real-world insights from Multiple Myeloma research applications
MM Research Signals (Observed)
Validated findings from Multiple Myeloma research applications
WIWFM Confidence (BRAF V600E)
Will-It-Work-For-Me confidence range for BRAF V600E variants in MM research applications
Efficacy Score Range
Efficacy score range (0.17-0.26) for MM variants with consistent performance
Fusion Coverage Usage
Fusion profile used only when AlphaMissense coverage exists - deterministic approach
Key Insight
Real-world MM research shows consistent confidence patterns and efficacy ranges, with fusion coverage providing comprehensive variant analysis.
CrisPRO.ai Delta Scoring Performance
Live genome-scale language model scoring with transcript-aware, multi-scale analysis
Strong Disruption Threshold
Delta score ≤ -3 indicates high functional disruption and resistance risk
Context Window Size
Optimal genomic context window (8,192 nt) for signal-to-noise balance
Multi-Scale Consistency
Estimated consistency across 1k/2k/4k/8k windows for high-confidence calls
Key Insight
Pathway Aggregation Results
RAS/MAPK and TP53 pathway disruption quantification for therapy response prediction
RAS/MAPK Pathway Coverage
Estimated coverage of KRAS/NRAS/BRAF variants in pathway aggregation
TP53 Cooperation Rate
Estimated frequency of TP53 alterations as cooperating hits in MM
Prediction Accuracy
Estimated sensitivity vs resistance prediction accuracy in validation cohort
Key Insight
Two-Hit Hypothesis (MM)
Multiple Myeloma follows a two-hit model with driver and cooperating alterations
MAPK Driver Frequency
Estimated frequency of MAPK pathway activation (BRAF/NRAS/KRAS)
TP53/17p Cooperation
Estimated frequency of TP53/17p alterations as cooperating hits
MYC Amplification
Estimated frequency of MYC amplification as cooperating alteration
Key Insight
Multiple Myeloma follows a clear two-hit model with MAPK pathway activation as the primary driver, often cooperating with TP53/17p alterations.
Clinical Trial Shortlist Compression
Efficiency gains in clinical trial matching for MM patients
Trial Shortlist Reduction
Reduction from 50+ to ~5-12 relevant trials
Time to Shortlist
Minutes to generate trial shortlist
Key Insight
Clinical trial matching efficiency dramatically improves with AI-powered shortlisting, reducing manual review time from hours to minutes.
Clinical Trial Shortlist Compression
Efficiency gains in clinical trial matching for MM patients
Trial Shortlist Reduction
Reduction from 50+ to ~5-12 relevant trials with AI-powered shortlisting
Time to Shortlist
Minutes to generate trial shortlist with Likely/Potential/Unlikely labels
Key Insight
Clinical trial matching efficiency dramatically improves with AI-powered shortlisting, reducing manual review time from hours to minutes.
Live Technical Pipeline
Expert-grade CrisPRO.ai genome-scale language model with strict data hygiene and transparent error handling
Input Validation
Strict data hygiene with allele and coordinate validation
Genomic Context
Fetch and center genomic window from Ensembl
CrisPRO.ai Delta Scoring
Live genome-scale language model scoring
Impact Mapping
Map delta scores to functional impact levels
Pathway Aggregation
Sum impacts into RAS/MAPK and TP53 pathways
JSON Output
Structured results with full provenance
Input Validation
Strict data hygiene with allele and coordinate validation
Hard fail on errors; no mock data generation
Genomic Context
Fetch and center genomic window from Ensembl
Optimal signal-to-noise balance for CrisPRO.ai scoring
CrisPRO.ai Delta Scoring
Live genome-scale language model scoring
Transcript-aware, no canned lookups
Impact Mapping
Map delta scores to functional impact levels
Clinically relevant thresholds for resistance prediction
Pathway Aggregation
Sum impacts into RAS/MAPK and TP53 pathways
Clinically relevant pathway focus for MM
JSON Output
Structured results with full provenance
Complete audit trail and repeatability
Input Validation
Strict data hygiene with allele and coordinate validation
Hard fail on errors; no mock data generation
Genomic Context
Fetch and center genomic window from Ensembl
Optimal signal-to-noise balance for CrisPRO.ai scoring
CrisPRO.ai Delta Scoring
Live genome-scale language model scoring
Transcript-aware, no canned lookups
Impact Mapping
Map delta scores to functional impact levels
Clinically relevant thresholds for resistance prediction
Pathway Aggregation
Sum impacts into RAS/MAPK and TP53 pathways
Clinically relevant pathway focus for MM
JSON Output
Structured results with full provenance
Complete audit trail and repeatability
What Makes This Expert-Grade
Live, transcript-aware scoring with strict data hygiene
Live CrisPRO.ai Scoring
No canned lookups, real-time genome-scale language model
Strict Data Hygiene
Fail rather than fabricate; transparent error handling
Clinically Relevant
KRAS/NRAS/BRAF and TP53 pathway focus
Extensible Architecture
Fusion-ready with splice-aware checks and protein models
Platform Capabilities
Comprehensive tools for variant analysis, therapy guidance, and clinical trial matching
Variant Insight (VUS)
Four chips (Function, Regulatory, Essentiality, Chromatin) in plain language. Turn unknowns into readable signals with helper text and thresholds.
Key Features
- Function chip: Protein impact prediction
- Regulatory chip: Non-coding variant effects
- Essentiality chip: Gene essentiality scores
- Chromatin chip: Chromatin accessibility impact
Why It Matters
- Reduces intake ambiguity; sets up therapy/pathway reasoning.
What We Delivered
- Live chips with helpers and provenance; export-ready.
Efficacy Intelligence (S/P/E Fusion)
S/P/E fusion: Sequence (CrisPRO.ai) + Pathway (burden) + Evidence (ClinVar/literature) into ranked therapy classes with explainable confidence and citations.
Key Features
- S/P/E fusion framework with insight chips
- Ranked drug classes with confidence scores
- Evidence integration with ClinVar + literature
- Cohort overlays for real-world context
Why It Matters
- Explainable ranking with confidence and citations (RUO).
- Faster decisions: clear starting point backed by sources.
- Tier promotions when ClinVar-Strong + Pathway-Aligned co-occur.
What We Delivered
- Live efficacy table with score, confidence, tier, badges, rationale, citations, provenance.
- Evidence panel with export capabilities and run IDs for repeatability.
Pathway View
Top 3 MM pathways with one-line "why" and contribution bars; links to therapy alignment.
Key Features
- MAPK pathway (BRAF/NRAS/KRAS)
- TP53/DDR pathway
- Proteostasis/CRBN pathway
Why It Matters
- A fast biology story that justifies therapy choices.
What We Delivered
- Stable top-3 pathways with bars, one-liners, and provenance.
Toxicity Risk (Germline)
Simple caution chip to plan conservatively. Confidence and sources included (RUO).
Key Features
- Germline variant screening
- Drug metabolism variants
- Toxicity risk scoring
Why It Matters
- Flags potential sensitivity early; improves patient communication.
What We Delivered
- Caution chip with helper, confidence, sources, provenance.
CRISPR Readiness (Demo)
Feasibility, access, off-target preview, delivery notes (demo). 1M-token context enables richer prompts.
Key Features
- On-target feasibility scoring
- Off-target prediction
- Delivery optimization
- Guide RNA design
Why It Matters
- Faster, safer starts for design exploration (research-mode).
What We Delivered
- Safety-gated candidates, access chip, demo off-target/delivery notes with provenance.
Clinical Trials Co-Pilot
Fast shortlist with Likely/Potential/Unlikely and a shareable one-pager. Synonym/biomarker-aware search and structured eligibility.
Key Features
- Smart trial matching
- Eligibility assessment
- One-pager export
- Real-time updates
Why It Matters
- Reduces 50+ trials to ~5–12 in minutes; improves patient/board alignment.
What We Delivered
- Shortlist with labels and “why”; export with run ID and sources.
How to Read Results
Confidence (0–1) is a certainty hint; Evidence Tier is Supported/Consider/Insufficient.
Badges show strength (Guideline, RCT, ClinVar-Strong, Pathway-Aligned).
Fusion labeled when eligible; Baseline remains deterministic.