Spectral Unmixing: A modular and reproducible Python package for directed and blind spectral unmixing in multidimensional microscopy stacks

In the world of biological imaging, scientists use powerful microscopes to peer into the tiny, glowing machinery of living cells. To make these invisible structures visible, researchers tag them with special dyes called fluorophores, which light up in specific colors when hit by a laser. By using several different dyes at once, they can watch multiple parts of a cell interact in real time. However, a persistent problem has long clouded these observations: the colors often bleed into one another. Just as a red light might cast a faint pink shadow on a nearby white wall, the glow from one dye can spill over into the channel meant for a different dye. This contamination creates a false signal, making it difficult to tell exactly where a molecule is or how much of it is present. For decades, fixing this issue has been a fragmented effort, relying on a patchwork of manual adjustments, custom scripts written by individual labs, or specialized tools that are hard to share or verify. Without a standard way to clean up these mixed signals, the data from different studies remains difficult to compare, and the true picture of cellular life stays slightly out of focus.

A new open-source software package called spectral-unmixing aims to bring order to this chaos by offering a single, reliable tool for correcting these color mix-ups. The researchers behind this work built a modular system that can handle the messy reality of microscope data, regardless of the file format the machine produces. Instead of forcing scientists to manually reorganize their data before analysis, the software automatically arranges the information into a standard order, allowing the correction process to begin immediately. The package is designed to be both flexible and transparent. It can perform directed corrections, where the user knows exactly which dyes are interfering with each other, and it can also attempt blind unmixing, where the software figures out the mixing patterns on its own without prior instructions. Crucially, every time the software runs, it creates a digital record of exactly what settings were used and what numbers it calculated. This sidecar file ensures that anyone can look back at the work later and see precisely how the final image was derived, removing the guesswork that often plagues scientific analysis.

The team tested their tool using both computer-generated simulations and real data from actual microscopy experiments to see how well it performed. In simulations where the true answer was known, the software proved highly effective at cleaning up the noise. When correcting a specific channel in a two-color setup, the error in the final image dropped significantly, moving from a normalized value of roughly 0.029 down to about 0.003. The software also showed great promise when dealing with data that changes over time. In these scenarios, calculating the correction for each moment individually proved far superior to using a single average setting for the entire experiment. This approach reduced the error in estimating the mixing levels from approximately 0.099 to just 0.003. Furthermore, the tool demonstrated a unique ability to handle bidirectional corrections, where two dyes interfere with each other simultaneously. By mathematically reversing the mixing model, the software reduced errors in the reciprocal channels from a range of 0.022 to 0.037 down to about 0.004.

For more complex situations involving many different colors at once, the software employs a method known as blind unmixing, which separates the mixed signals without needing to know the exact properties of the dyes beforehand. Benchmarks on multichannel data showed that this method successfully reduced the unwanted dependence between channels while keeping the identity of each original fluorophore intact. The researchers also found that if a scientist knows the likely path of contamination—for instance, knowing that dye A often leaks into dye B's channel—they can use specific constraints to guide the software, offering a controllable alternative when the mixing pathways are understood. The results suggest that this package provides a robust platform for standardizing how scientists clean up their images. By making the process scriptable, reproducible, and accessible, the tool lowers the barrier to entry for high-quality analysis, allowing researchers to trust that the glowing patterns they see in their microscopes are a true reflection of the biology, not an artifact of the equipment.