Figure 4. Dynamic range and Z-factors of positive and negative controls in screening plates.

(A) Comparison of Z-factor of nuclear count and MYBPC3 and ACTN2 staining intensities for dimethyl sulfoxide (DMSO) vs. bortezomib (Bort; 0.1 μM) and doxorubicin (Doxo; 1 μM). The Z-factors (< 0) and the limited dynamic range of the nuclear count and sarcomere intensities prevent reliable separation of positive and negative controls. Error bars = SD. (B) While using deep learning (regardless of how the models are trained), the dynamic range and Z-factors (> 0.5) enable identification of the toxic controls from the DMSO condition. Error bars = SD. (C) Cardiotoxicity scores from all screening wells at three doses of DMSO, bortezomib, and doxorubicin controls are compared in various deep learning models. Regardless of how each deep learning model was trained, cardiotoxicity scores from all three models had strong correlation when applied to the screening plates (R² > 0.95).

Figure 4—figure supplement 1. Sarcomere intensity and nuclear count on screening plates at three doses.

(A) Signal intensity and sarcomere content in screening plates using antibodies against MYBPC3 and ACTN2. Nuclear count was measured using Hoechst stain. (B) Two-dimensional plots of MYBPC3 and ACTN2 signal intensity show a good correlation between the two sarcomeric stains. Dimethyl sulfoxide (DMSO) (0.1%), bortezomib (0.1 μM), and doxorubicin (1.0 μM) are used as internal controls.