Multi-Modal Integration
Why combination of orthogonal measurement dimensions beats any single endpoint for developmental neurotoxicity detection
Three Orthogonal Measurement Dimensions
PCA on 9 morphological features (N=64 organoids) reveals 3 independent dimensions capturing 99.4% of total variance.
PC1-Size
71.9%Overall organoid size, dominated by area, perimeter, and axis measurements. Captures proliferative capacity and growth rate differences.
Top loadings
Discriminates
PC2-Asymmetry
19.4%Shape asymmetry and directional growth bias. Captures cytoskeletal polarity and tissue architecture disruption.
Top loadings
Discriminates
PC3-Roundness
8.2%Tissue organization quality and structural integrity. Captures self-assembly quality and boundary regularity.
Top loadings
Discriminates
Single vs. Multi-Modal Detection
Each PCA composite detects a different subset of clones. Combining all three ensures no disease model escapes detection.
False Positive Rate Reduction
Requiring concordance across independent composites exponentially reduces false positive rates.
15.0%
Single composite
1 composite required
2.3%
Two concordant
2 composites required
0.3%
Three concordant
3 composites required
Under independence assumption: P(concordant false positive) = (single FPR)^n. Single composite FPR \u2248 15% (Bonferroni-adjusted threshold). Two concordant: 0.15² = 2.3%. Three concordant: 0.15³ = 0.3%.
Pass / Flag / Fail Classifier
Toggle between single-modal (PC1 only) and multi-modal (all 3 PCs) to see how combinatorial analysis improves classification confidence.
Wildtype Control
0 composites flagged
TUBA1A Mutant
2 composites flagged
TUBB2A Mutant
3 composites flagged
TH Deficient
2 composites flagged
Principle Validated in Regulatory Science
The multi-endpoint integration strategy is not novel to ConductScreen — it is the established approach in regulatory toxicology. EPA's integrated DNT battery achieved 93% sensitivity by combining multiple NAM endpoints (Carstens et al. 2022, Alternatives to Laboratory Animals). ConductScreen applies this same principle within a single automated workflow: multiple orthogonal morphological composites provide cross-modal confirmation that no single endpoint can achieve alone.
Phase 2 Expansion
Phase 2 adds two additional orthogonal measurement dimensions, further reducing false positive rates:
Temporal Module
Growth trajectory composites add time-series dynamics as a 4th orthogonal dimension
Boundary Module
Neurite topology composites add structural connectivity as a 5th orthogonal dimension
With 5 orthogonal composites: P(concordant false positive) = 0.15&sup5; = 0.0076% — approaching zero false positive risk for ≥2 concordant composites.