in silico
Phase 1 : In Silico Validation
- Objectif
- To computationally falsify the core causal chain (stiffness → YAP/TAZ nuclear translocation → Notch inhibition) using existing single-cell RNA-seq and imaging datasets, and to validate the feasibility of unsupervised clustering for separating stiffness conditions.
- Coût estimé
- €500-2000
- Durée estimée
- 4-6 weeks
- Critères de réussite
- Spearman correlation between YAP/TAZ and Notch activity scores in public scRNA-seq data · ρ < -0.3 (negative correlation) with p < 0.05 · (Compute using cor.test() in R or scipy.stats.spearmanr in Python)
- ABM prediction: N/C ratio increase at 50 kPa vs 0.5 kPa · ≥1.5-fold increase (from baseline 0.6 to >0.9) after 48h · (Mean N/C ratio from ABM output across 10 runs)
- ABM prediction: Notch reporter decrease at 50 kPa vs 0.5 kPa · ≥20% decrease (from baseline 100 a.u. to ≤80 a.u.) · (Mean Notch reporter intensity from ABM output across 10 runs)
- Cluster centroid correlation in synthetic images · Pearson r < -0.6 with p < 0.01 · (Pearson correlation test on cluster centroids from PhenoGraph)
- On continue si
- All four success criteria are met (negative correlation in scRNA-seq, ABM predicts ≥1.5-fold N/C increase and ≥20% Notch decrease, cluster centroid r < -0.6).
- On arrête si
- Any criterion fails (e.g., scRNA-seq shows positive or no correlation, ABM predicts <1.5-fold N/C increase, or clustering yields <3 clusters or r > -0.6).
- On réoriente si
- If scRNA-seq shows negative correlation but ABM fails, refine ABM parameters (e.g., increase YAP/TAZ shuttling rate) and re-run. If clustering fails, try alternative clustering methods (e.g., HDBSCAN, Leiden) or reduce feature set to top 50 features by variance.
- Risques
- Public scRNA-seq datasets do not include matrix stiffness as a variable.Probabilité : mediumParade : Use datasets with YAP/TAZ perturbation (e.g., Verteporfin treatment) as a proxy for stiffness variation. Alternatively, use spatial transcriptomics data (e.g., MERFISH) where stiffness can be inferred from cell density.
- ABM simulations are too simplistic to capture non-linear dynamics.Probabilité : mediumParade : Validate ABM against known experimental results from the literature (e.g., DOI:10.1038/ncomms15206). Increase model complexity incrementally (e.g., add cell cycle, apoptosis).
- Synthetic images do not produce realistic morphological features.Probabilité : lowParade : Use a generative adversarial network (GAN) trained on real spheroid images (e.g., from IDR) to generate more realistic synthetic data. Alternatively, use a simple parametric model (e.g., ellipsoid nuclei with Gaussian noise).