Fig 1.
Structural changes in C2C12 myoblasts treated with colchicine.
Composite fluorescence microscopy images captured after a 72 h treatment with DMSO 0.1% as vehicle control (A), or COLC at various concentrations: 10 nM (B), 3 µ M (C), 30 µ M (D). 20X magnification, scale bar of 50 µ M for reference. Channel color mapping: blue - DNA, red - Mitochondria, green - ER, magenta - Actin, yellow - RNA. Contrast adjusted on DNA channel for improved readability. (E) The density of nuclear area (µm2) of control and Colchicine (100 nM) treated myoblasts and (F) myotubes.
Fig 2.
Median cell counts per well (A) and median cell viability (B) plotted against 8 concentrations for the 30 compounds of the Toxifate library in C2C12 myoblasts treatments.
Results are shown as median value ± standard error from three biological replicates and four technical replicates, n = 12 (3 different plates plated on different days). Statistical significance was assessed by independent T-tests against the mean of vehicle controls for each motherplate. Meaning of symbols: * : p-value < 0.05, **: p-value < 0.01, ***: p-value < 0.001, no symbol: ns.
Fig 3.
Prediction performance of cell count prediction models built on C2C12 myoblasts and myotubes on Harmony data.
Test set determination coefficients (R2) of predicted cell counts against observed cell counts, from Random Forest models built for 30 compounds of the Toxifate library of suspected myotoxicants.
Fig 4.
Prediction performance of cell viability prediction models built on C2C12 myoblasts and myotubes on Harmony data.
Test set determination coefficients (R2) of predicted cell viability against observed viability, from Random Forest models built for 30 compounds of the Toxifate library. Luminescence data scaled against median luminescence of healthy DMSO controls (represented by 100% luminescence).
Fig 5.
Heatmap and hierarchical clustering of top 30 Random Forest features importance for each model built on individual drugs for viability prediction in myoblasts.
Features are ranked in descending order by their median rank of importance across all models (most important features are at the top of the heatmap). Clustering is performed using single-linkage and correlation metric. The lower half of the heatmap containing less prominent features was cropped for readability.
Table 1.
Summary of RF feature importance for the top 10 properties of the Random Forest model for viability prediction built on C2C12 myoblasts with the Harmony dataset.
Fig 6.
Heatmap and hierarchical clustering of treatment-level morphological profiles for C2C12 myoblasts.
Only profiles with induction > 0.2 are represented. Hierarchical clustering of treatments and features computed with average-linkage on Pearson correlations. Z-scores clipped to ± 8 MAD for increased readability and color scaling.
Table 2.
Toxifate drug panel of myotoxicants.
Fig 7.
Outline of the CPA protocol applied to C2C12 cells.
Color coding, Blue: cell culture and staining, Yellow: high-content microscopy acquisition, Red: data analysis.