in silico
Phase 1 : In Silico Validation
- Objectif
- To computationally confirm that the proposed synergy detection method (PROGENy with interaction term) can distinguish non-additive transcriptomic effects from additive ones using existing public datasets, and to estimate the expected effect size and statistical power before committing to animal experiments.
- Coût estimé
- €500-2000
- Durée estimée
- 4-6 weeks
- Critères de réussite
- ΔNES for HGF/MET pathway in simulated or real combination data · ≥ 1.5 SD above null distribution mean (from 1000 permutations) · (Compute ΔNES = NES_combination - (NES_biomaterial + NES_hASC); compare to null distribution via z-score)
- Statistical power to detect ΔNES ≥ 1.5 SD · Power ≥ 0.8 at α=0.05 with n=6 per group · (G*Power one-sample z-test: effect size d = ΔNES / SD_null, input n=6, α=0.05, output power)
- Sensitivity of ΔNES to gene set collection · ΔNES for HGF/MET pathway remains ≥ 1.0 SD in at least 2 out of 3 collections (PROGENy, Hallmark, KEGG) · (Repeat PROGENy analysis with Hallmark and KEGG gene sets; compute ΔNES for each)
- On continue si
- ΔNES for HGF/MET pathway ≥ 1.5 SD above null mean AND power ≥ 0.8 with n=6 per group AND sensitivity confirmed (ΔNES ≥ 1.0 SD in ≥2 collections).
- On arrête si
- ΔNES < 1.0 SD above null mean OR power < 0.6 with n=6 per group OR sensitivity fails (ΔNES < 0.5 SD in all collections).
- On réoriente si
- ΔNES between 1.0-1.5 SD OR power between 0.6-0.8: increase sample size to n=8-10 per group in Phase 2, or test alternative pathway (e.g., angiogenesis) with higher ΔNES.
- Risques
- No public combination dataset exists; simulated data may not reflect true biology.Probabilité : highParade : Use linear mixing of individual treatment expression matrices with noise (N(0, σ²)) to simulate additive and synergistic profiles. Validate simulation by comparing to known drug combination synergy datasets (e.g., from DrugComb portal).
- PROGENy interaction term implementation is not straightforward in R.Probabilité : mediumParade : Use the “progeny” function with “interaction=TRUE” parameter (if available in v1.24.0). Alternatively, compute NES separately for each group and use linear model: NES ~ treatment_A * treatment_B, extracting the interaction coefficient.
- Null distribution from label permutation may be under-dispersed due to small sample size in public datasets.Probabilité : mediumParade : Use bootstrapping (resampling with replacement) in addition to permutation to estimate null distribution variability. Increase permutation number to 5000 if needed.