I appreciated the novel approach to reconstructing super-resolved images of cells in flow, and the demonstrations in real biological samples with E. gracilis lipid droplets and CD8+ T-cell mitochondria.

I wondered whether you have considered testing sensitivity experimentally, beyond the simulated noise analysis in Fig. 2e. For example, beads spanning a range of brightness could show how resolution, and the detection limit, change as signal drops. Ideally these would be imaged in flow, since the beads in Fig. 3a–d were moved on a stage. Relatedly, in Fig. 3d the full width at half maximum (FWHM) differs slightly between the flow axis (155 nm) and the transverse axis (149 nm). Do you expect that difference to depend on signal intensity, given that the flow axis is measured while the transverse axis is reconstructed by the network?

Finally, I’m curious whether you think this approach could extend to lower-signal targets, such as lipid droplets in yeast. Yeast droplets are typically smaller and fewer per cell than those in E. gracilis, so each droplet would likely give much less signal. Knowing the approximate photon counts per droplet in your E. gracilis data would help in judging whether such samples are within range.

I used Claude (Opus 5.5) to help craft the language in this feedback, but verified that the content is accurate.